Mirrored_Repos/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI synced 2024-08-30 20:32:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

resolve conflicts with main

Browse Source
This commit is contained in:
Lincoln Stein 2023-07-17 17:10:57 -04:00
commit 65ed43afb9
483 changed files with 20162 additions and 19583 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
.dockerignore
View File

@ -1,25 +1,9 @@
# use this file as a whitelist
*
!invokeai
!ldm
!pyproject.toml
!docker/docker-entrypoint.sh
!LICENSE
# ignore frontend/web but whitelist dist
invokeai/frontend/web/
!invokeai/frontend/web/dist/
# ignore invokeai/assets but whitelist invokeai/assets/web
invokeai/assets/
!invokeai/assets/web/
# Guard against pulling in any models that might exist in the directory tree
**/*.pt*
**/*.ckpt
# Byte-compiled / optimized / DLL files
**/__pycache__/
**/*.py[cod]
# Distribution / packaging
**/*.egg-info/
**/*.egg
**/node_modules
**/__pycache__
**/*.egg-info

4
.github/CODEOWNERS vendored
View File

@ -6,7 +6,7 @@
/mkdocs.yml @lstein @blessedcoolant
# nodes
/invokeai/app/ @Kyle0654 @blessedcoolant
/invokeai/app/ @Kyle0654 @blessedcoolant @psychedelicious @brandonrising
# installation and configuration
/pyproject.toml @lstein @blessedcoolant
@ -22,7 +22,7 @@
/invokeai/backend @blessedcoolant @psychedelicious @lstein @maryhipp
# generation, model management, postprocessing
/invokeai/backend @damian0815 @lstein @blessedcoolant @jpphoto @gregghelt2 @StAlKeR7779
/invokeai/backend @damian0815 @lstein @blessedcoolant @gregghelt2 @StAlKeR7779 @brandonrising
# front ends
/invokeai/frontend/CLI @lstein

83
.github/workflows/build-container.yml vendored
View File

@ -3,17 +3,15 @@ on:
push:
branches:
- 'main'
- 'update/ci/docker/*'
- 'update/docker/*'
- 'dev/ci/docker/*'
- 'dev/docker/*'
paths:
- 'pyproject.toml'
- '.dockerignore'
- 'invokeai/**'
- 'docker/Dockerfile'
- 'docker/docker-entrypoint.sh'
- 'workflows/build-container.yml'
tags:
- 'v*.*.*'
- 'v*'
workflow_dispatch:
permissions:
@ -26,23 +24,27 @@ jobs:
strategy:
fail-fast: false
matrix:
flavor:
- rocm
- cuda
- cpu
include:
- flavor: rocm
pip-extra-index-url: 'https://download.pytorch.org/whl/rocm5.2'
- flavor: cuda
pip-extra-index-url: ''
- flavor: cpu
pip-extra-index-url: 'https://download.pytorch.org/whl/cpu'
gpu-driver:
- cuda
- cpu
- rocm
runs-on: ubuntu-latest
name: ${{ matrix.flavor }}
name: ${{ matrix.gpu-driver }}
env:
PLATFORMS: 'linux/amd64,linux/arm64'
DOCKERFILE: 'docker/Dockerfile'
# torch/arm64 does not support GPU currently, so arm64 builds
# would not be GPU-accelerated.
# re-enable arm64 if there is sufficient demand.
# PLATFORMS: 'linux/amd64,linux/arm64'
PLATFORMS: 'linux/amd64'
steps:
- name: Free up more disk space on the runner
# https://github.com/actions/runner-images/issues/2840#issuecomment-1284059930
run: |
sudo rm -rf /usr/share/dotnet
sudo rm -rf "$AGENT_TOOLSDIRECTORY"
sudo swapoff /mnt/swapfile
sudo rm -rf /mnt/swapfile
- name: Checkout
uses: actions/checkout@v3
@ -53,7 +55,7 @@ jobs:
github-token: ${{ secrets.GITHUB_TOKEN }}
images: |
ghcr.io/${{ github.repository }}
${{ vars.DOCKERHUB_REPOSITORY }}
${{ env.DOCKERHUB_REPOSITORY }}
tags: |
type=ref,event=branch
type=ref,event=tag
@ -62,8 +64,8 @@ jobs:
type=pep440,pattern={{major}}
type=sha,enable=true,prefix=sha-,format=short
flavor: |
latest=${{ matrix.flavor == 'cuda' && github.ref == 'refs/heads/main' }}
suffix=-${{ matrix.flavor }},onlatest=false
latest=${{ matrix.gpu-driver == 'cuda' && github.ref == 'refs/heads/main' }}
suffix=-${{ matrix.gpu-driver }},onlatest=false
- name: Set up QEMU
uses: docker/setup-qemu-action@v2
@ -81,34 +83,33 @@ jobs:
username: ${{ github.repository_owner }}
password: ${{ secrets.GITHUB_TOKEN }}
- name: Login to Docker Hub
if: github.event_name != 'pull_request' && vars.DOCKERHUB_REPOSITORY != ''
uses: docker/login-action@v2
with:
username: ${{ secrets.DOCKERHUB_USERNAME }}
password: ${{ secrets.DOCKERHUB_TOKEN }}
# - name: Login to Docker Hub
# if: github.event_name != 'pull_request' && vars.DOCKERHUB_REPOSITORY != ''
# uses: docker/login-action@v2
# with:
# username: ${{ secrets.DOCKERHUB_USERNAME }}
# password: ${{ secrets.DOCKERHUB_TOKEN }}
- name: Build container
id: docker_build
uses: docker/build-push-action@v4
with:
context: .
file: ${{ env.DOCKERFILE }}
file: docker/Dockerfile
platforms: ${{ env.PLATFORMS }}
push: ${{ github.ref == 'refs/heads/main' || github.ref_type == 'tag' }}
tags: ${{ steps.meta.outputs.tags }}
labels: ${{ steps.meta.outputs.labels }}
build-args: PIP_EXTRA_INDEX_URL=${{ matrix.pip-extra-index-url }}
cache-from: |
type=gha,scope=${{ github.ref_name }}-${{ matrix.flavor }}
type=gha,scope=main-${{ matrix.flavor }}
cache-to: type=gha,mode=max,scope=${{ github.ref_name }}-${{ matrix.flavor }}
type=gha,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}
type=gha,scope=main-${{ matrix.gpu-driver }}
cache-to: type=gha,mode=max,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}
- name: Docker Hub Description
if: github.ref == 'refs/heads/main' || github.ref == 'refs/tags/*' && vars.DOCKERHUB_REPOSITORY != ''
uses: peter-evans/dockerhub-description@v3
with:
username: ${{ secrets.DOCKERHUB_USERNAME }}
password: ${{ secrets.DOCKERHUB_TOKEN }}
repository: ${{ vars.DOCKERHUB_REPOSITORY }}
short-description: ${{ github.event.repository.description }}
# - name: Docker Hub Description
# if: github.ref == 'refs/heads/main' || github.ref == 'refs/tags/*' && vars.DOCKERHUB_REPOSITORY != ''
# uses: peter-evans/dockerhub-description@v3
# with:
# username: ${{ secrets.DOCKERHUB_USERNAME }}
# password: ${{ secrets.DOCKERHUB_TOKEN }}
# repository: ${{ vars.DOCKERHUB_REPOSITORY }}
# short-description: ${{ github.event.repository.description }}

6
.gitignore vendored
View File

@ -34,7 +34,7 @@ __pycache__/
.Python
build/
develop-eggs/
dist/
# dist/
downloads/
eggs/
.eggs/
@ -79,6 +79,7 @@ cov.xml
.pytest.ini
cover/
junit/
notes/
# Translations
*.mo
@ -201,7 +202,8 @@ checkpoints
# If it's a Mac
.DS_Store
invokeai/frontend/web/dist/*
invokeai/frontend/yarn.lock
invokeai/frontend/node_modules
# Let the frontend manage its own gitignore
!invokeai/frontend/web/*

189
LICENSE
View File

@ -1,21 +1,176 @@
MIT License
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
Copyright (c) 2022 InvokeAI Team
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
1. Definitions.
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.

226
README.md
View File

@ -1,8 +1,11 @@
<div align="center">
![project logo](https://github.com/invoke-ai/InvokeAI/raw/main/docs/assets/invoke_ai_banner.png)
![project hero](https://github.com/invoke-ai/InvokeAI/assets/31807370/1a917d94-e099-4fa1-a70f-7dd8d0691018)
# Invoke AI - Generative AI for Professional Creatives
## Professional Creative Tools for Stable Diffusion, Custom-Trained Models, and more.
To learn more about Invoke AI, get started instantly, or implement our Business solutions, visit [invoke.ai](https://invoke.ai)
# InvokeAI: A Stable Diffusion Toolkit
[![discord badge]][discord link]
@ -33,32 +36,32 @@
</div>
_**Note: The UI is not fully functional on `main`. If you need a stable UI based on `main`, use the `pre-nodes` tag while we [migrate to a new backend](https://github.com/invoke-ai/InvokeAI/discussions/3246).**_
_**Note: This is an alpha release. Bugs are expected and not all
features are fully implemented. Please use the GitHub [Issues
pages](https://github.com/invoke-ai/InvokeAI/issues?q=is%3Aissue+is%3Aopen)
to report unexpected problems. Also note that InvokeAI root directory
which contains models, outputs and configuration files, has changed
between the 2.x and 3.x release. If you wish to use your v2.3 root
directory with v3.0, please follow the directions in [Migrating a 2.3
root directory to 3.0](#migrating-to-3).**_
InvokeAI is a leading creative engine built to empower professionals and enthusiasts alike. Generate and create stunning visual media using the latest AI-driven technologies. InvokeAI offers an industry leading Web Interface, interactive Command Line Interface, and also serves as the foundation for multiple commercial products.
InvokeAI is a leading creative engine built to empower professionals
and enthusiasts alike. Generate and create stunning visual media using
the latest AI-driven technologies. InvokeAI offers an industry leading
Web Interface, interactive Command Line Interface, and also serves as
the foundation for multiple commercial products.
**Quick links**: [[How to Install](https://invoke-ai.github.io/InvokeAI/#installation)] [<a href="https://discord.gg/ZmtBAhwWhy">Discord Server</a>] [<a href="https://invoke-ai.github.io/InvokeAI/">Documentation and Tutorials</a>] [<a href="https://github.com/invoke-ai/InvokeAI/">Code and Downloads</a>] [<a href="https://github.com/invoke-ai/InvokeAI/issues">Bug Reports</a>] [<a href="https://github.com/invoke-ai/InvokeAI/discussions">Discussion, Ideas & Q&A</a>]
_Note: InvokeAI is rapidly evolving. Please use the
[Issues](https://github.com/invoke-ai/InvokeAI/issues) tab to report bugs and make feature
requests. Be sure to use the provided templates. They will help us diagnose issues faster._
## FOR DEVELOPERS - MIGRATING TO THE 3.0.0 MODELS FORMAT
The models directory and models.yaml have changed. To migrate to the
new layout, please follow this recipe:
1. Run `python scripts/migrate_models_to_3.0.py <path_to_root_directory>
2. This will create a new models directory named `models-3.0` and a
new config directory named `models.yaml-3.0`, both in the current
working directory. If you prefer to name them something else, pass
the `--dest-directory` and/or `--dest-yaml` arguments.
3. Check that the new models directory and yaml file look ok.
4. Replace the existing directory and file, keeping backup copies just in
case.
**Quick links**: [[How to
Install](https://invoke-ai.github.io/InvokeAI/#installation)] [<a
href="https://discord.gg/ZmtBAhwWhy">Discord Server</a>] [<a
href="https://invoke-ai.github.io/InvokeAI/">Documentation and
Tutorials</a>] [<a
href="https://github.com/invoke-ai/InvokeAI/">Code and
Downloads</a>] [<a
href="https://github.com/invoke-ai/InvokeAI/issues">Bug Reports</a>]
[<a
href="https://github.com/invoke-ai/InvokeAI/discussions">Discussion,
Ideas & Q&A</a>]
<div align="center">
@ -68,22 +71,30 @@ case.
## Table of Contents
1. [Quick Start](#getting-started-with-invokeai)
2. [Installation](#detailed-installation-instructions)
3. [Hardware Requirements](#hardware-requirements)
4. [Features](#features)
5. [Latest Changes](#latest-changes)
6. [Troubleshooting](#troubleshooting)
7. [Contributing](#contributing)
8. [Contributors](#contributors)
9. [Support](#support)
10. [Further Reading](#further-reading)
Table of Contents 📝
## Getting Started with InvokeAI
**Getting Started**
1. 🏁 [Quick Start](#quick-start)
3. 🖥️ [Hardware Requirements](#hardware-requirements)
**More About Invoke**
1. 🌟 [Features](#features)
2. 📣 [Latest Changes](#latest-changes)
3. 🛠️ [Troubleshooting](#troubleshooting)
**Supporting the Project**
1. 🤝 [Contributing](#contributing)
2. 👥 [Contributors](#contributors)
3. 💕 [Support](#support)
## Quick Start
For full installation and upgrade instructions, please see:
[InvokeAI Installation Overview](https://invoke-ai.github.io/InvokeAI/installation/)
If upgrading from version 2.3, please read [Migrating a 2.3 root
directory to 3.0](#migrating-to-3) first.
### Automatic Installer (suggested for 1st time users)
1. Go to the bottom of the [Latest Release Page](https://github.com/invoke-ai/InvokeAI/releases/latest)
@ -92,9 +103,8 @@ For full installation and upgrade instructions, please see:
3. Unzip the file.
4. If you are on Windows, double-click on the `install.bat` script. On
macOS, open a Terminal window, drag the file `install.sh` from Finder
into the Terminal, and press return. On Linux, run `install.sh`.
4. **Windows:** double-click on the `install.bat` script. **macOS:** Open a Terminal window, drag the file `install.sh` from Finder
into the Terminal, and press return. **Linux:** run `install.sh`.
5. You'll be asked to confirm the location of the folder in which
to install InvokeAI and its image generation model files. Pick a
@ -120,10 +130,12 @@ and go to http://localhost:9090.
10. Type `banana sushi` in the box on the top left and click `Invoke`
### Command-Line Installation (for users familiar with Terminals)
### Command-Line Installation (for developers and users familiar with Terminals)
You must have Python 3.9 or 3.10 installed on your machine. Earlier or later versions are
not supported.
You must have Python 3.9 or 3.10 installed on your machine. Earlier or
later versions are not supported.
Node.js also needs to be installed along with yarn (can be installed with
the command `npm install -g yarn` if needed)
1. Open a command-line window on your machine. The PowerShell is recommended for Windows.
2. Create a directory to install InvokeAI into. You'll need at least 15 GB of free space:
@ -187,16 +199,23 @@ not supported.
7. Launch the web server (do it every time you run InvokeAI):
```terminal
invokeai --web
invokeai-web
```
8. Point your browser to http://localhost:9090 to bring up the web interface.
9. Type `banana sushi` in the box on the top left and click `Invoke`.
8. Build Node.js assets
```terminal
cd invokeai/frontend/web/
yarn vite build
```
9. Point your browser to http://localhost:9090 to bring up the web interface.
10. Type `banana sushi` in the box on the top left and click `Invoke`.
Be sure to activate the virtual environment each time before re-launching InvokeAI,
using `source .venv/bin/activate` or `.venv\Scripts\activate`.
### Detailed Installation Instructions
## Detailed Installation Instructions
This fork is supported across Linux, Windows and Macintosh. Linux
users can use either an Nvidia-based card (with CUDA support) or an
@ -204,6 +223,87 @@ AMD card (using the ROCm driver). For full installation and upgrade
instructions, please see:
[InvokeAI Installation Overview](https://invoke-ai.github.io/InvokeAI/installation/INSTALL_SOURCE/)
<a name="migrating-to-3"></a>
### Migrating a v2.3 InvokeAI root directory
The InvokeAI root directory is where the InvokeAI startup file,
installed models, and generated images are stored. It is ordinarily
named `invokeai` and located in your home directory. The contents and
layout of this directory has changed between versions 2.3 and 3.0 and
cannot be used directly.
We currently recommend that you use the installer to create a new root
directory named differently from the 2.3 one, e.g. `invokeai-3` and
then use a migration script to copy your 2.3 models into the new
location. However, if you choose, you can upgrade this directory in
place. This section gives both recipes.
#### Creating a new root directory and migrating old models
This is the safer recipe because it leaves your old root directory in
place to fall back on.
1. Follow the instructions above to create and install InvokeAI in a
directory that has a different name from the 2.3 invokeai directory.
In this example, we will use "invokeai-3"
2. When you are prompted to select models to install, select a minimal
set of models, such as stable-diffusion-v1.5 only.
3. After installation is complete launch `invokeai.sh` (Linux/Mac) or
`invokeai.bat` and select option 8 "Open the developers console". This
will take you to the command line.
4. Issue the command `invokeai-migrate3 --from /path/to/v2.3-root --to
/path/to/invokeai-3-root`. Provide the correct `--from` and `--to`
paths for your v2.3 and v3.0 root directories respectively.
This will copy and convert your old models from 2.3 format to 3.0
format and create a new `models` directory in the 3.0 directory. The
old models directory (which contains the models selected at install
time) will be renamed `models.orig` and can be deleted once you have
confirmed that the migration was successful.
#### Migrating in place
For the adventurous, you may do an in-place upgrade from 2.3 to 3.0
without touching the command line. The recipe is as follows>
1. Launch the InvokeAI launcher script in your current v2.3 root directory.
2. Select option [9] "Update InvokeAI" to bring up the updater dialog.
3a. During the alpha release phase, select option [3] and manually
enter the tag name `v3.0.0+a2`.
3b. Once 3.0 is released, select option [1] to upgrade to the latest release.
4. Once the upgrade is finished you will be returned to the launcher
menu. Select option [7] "Re-run the configure script to fix a broken
install or to complete a major upgrade".
This will run the configure script against the v2.3 directory and
update it to the 3.0 format. The following files will be replaced:
- The invokeai.init file, replaced by invokeai.yaml
- The models directory
- The configs/models.yaml model index
The original versions of these files will be saved with the suffix
".orig" appended to the end. Once you have confirmed that the upgrade
worked, you can safely remove these files. Alternatively you can
restore a working v2.3 directory by removing the new files and
restoring the ".orig" files' original names.
#### Migration Caveats
The migration script will migrate your invokeai settings and models,
including textual inversion models, LoRAs and merges that you may have
installed previously. However it does **not** migrate the generated
images stored in your 2.3-format outputs directory. The released
version of 3.0 is expected to have an interface for importing an
entire directory of image files as a batch.
## Hardware Requirements
InvokeAI is supported across Linux, Windows and macOS. Linux
@ -222,13 +322,9 @@ We do not recommend the GTX 1650 or 1660 series video cards. They are
unable to run in half-precision mode and do not have sufficient VRAM
to render 512x512 images.
### Memory
**Memory** - At least 12 GB Main Memory RAM.
- At least 12 GB Main Memory RAM.
### Disk
- At least 12 GB of free disk space for the machine learning model, Python, and all its dependencies.
**Disk** - At least 12 GB of free disk space for the machine learning model, Python, and all its dependencies.
## Features
@ -242,28 +338,24 @@ InvokeAI offers a locally hosted Web Server & React Frontend, with an industry l
The Unified Canvas is a fully integrated canvas implementation with support for all core generation capabilities, in/outpainting, brush tools, and more. This creative tool unlocks the capability for artists to create with AI as a creative collaborator, and can be used to augment AI-generated imagery, sketches, photography, renders, and more.
### *Advanced Prompt Syntax*
### *Node Architecture & Editor (Beta)*
InvokeAI's advanced prompt syntax allows for token weighting, cross-attention control, and prompt blending, allowing for fine-tuned tweaking of your invocations and exploration of the latent space.
Invoke AI's backend is built on a graph-based execution architecture. This allows for customizable generation pipelines to be developed by professional users looking to create specific workflows to support their production use-cases, and will be extended in the future with additional capabilities.
### *Command Line Interface*
### *Board & Gallery Management*
For users utilizing a terminal-based environment, or who want to take advantage of CLI features, InvokeAI offers an extensive and actively supported command-line interface that provides the full suite of generation functionality available in the tool.
Invoke AI provides an organized gallery system for easily storing, accessing, and remixing your content in the Invoke workspace. Images can be dragged/dropped onto any Image-base UI element in the application, and rich metadata within the Image allows for easy recall of key prompts or settings used in your workflow.
### Other features
- *Support for both ckpt and diffusers models*
- *SD 2.0, 2.1 support*
- *Noise Control & Tresholding*
- *Popular Sampler Support*
- *Upscaling & Face Restoration Tools*
- *Upscaling Tools*
- *Embedding Manager & Support*
- *Model Manager & Support*
### Coming Soon
- *Node-Based Architecture & UI*
- And more...
- *Node-Based Architecture*
- *Node-Based Plug-&-Play UI (Beta)*
- *SDXL Support* (Coming soon)
### Latest Changes
@ -271,7 +363,7 @@ For our latest changes, view our [Release
Notes](https://github.com/invoke-ai/InvokeAI/releases) and the
[CHANGELOG](docs/CHANGELOG.md).
## Troubleshooting
### Troubleshooting
Please check out our **[Q&A](https://invoke-ai.github.io/InvokeAI/help/TROUBLESHOOT/#faq)** to get solutions for common installation
problems and other issues.
@ -301,8 +393,6 @@ This fork is a combined effort of various people from across the world.
[Check out the list of all these amazing people](https://invoke-ai.github.io/InvokeAI/other/CONTRIBUTORS/). We thank them for
their time, hard work and effort.
Thanks to [Weblate](https://weblate.org/) for generously providing translation services to this project.
### Support
For support, please use this repository's GitHub Issues tracking service, or join the Discord.

13
docker/.env.sample Normal file
View File

@ -0,0 +1,13 @@
## Make a copy of this file named `.env` and fill in the values below.
## Any environment variables supported by InvokeAI can be specified here.
# INVOKEAI_ROOT is the path to a path on the local filesystem where InvokeAI will store data.
# Outputs will also be stored here by default.
# This **must** be an absolute path.
INVOKEAI_ROOT=
HUGGINGFACE_TOKEN=
## optional variables specific to the docker setup
# GPU_DRIVER=cuda
# CONTAINER_UID=1000

206
docker/Dockerfile
View File

@ -1,107 +1,129 @@
# syntax=docker/dockerfile:1
# syntax=docker/dockerfile:1.4
ARG PYTHON_VERSION=3.9
##################
## base image ##
##################
FROM --platform=${TARGETPLATFORM} python:${PYTHON_VERSION}-slim AS python-base
## Builder stage
LABEL org.opencontainers.image.authors="mauwii@outlook.de"
FROM library/ubuntu:22.04 AS builder
# Prepare apt for buildkit cache
RUN rm -f /etc/apt/apt.conf.d/docker-clean \
&& echo 'Binary::apt::APT::Keep-Downloaded-Packages "true";' >/etc/apt/apt.conf.d/keep-cache
ARG DEBIAN_FRONTEND=noninteractive
RUN rm -f /etc/apt/apt.conf.d/docker-clean; echo 'Binary::apt::APT::Keep-Downloaded-Packages "true";' > /etc/apt/apt.conf.d/keep-cache
RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
--mount=type=cache,target=/var/lib/apt,sharing=locked \
apt update && apt-get install -y \
git \
python3.10-venv \
python3-pip \
build-essential
# Install dependencies
RUN \
--mount=type=cache,target=/var/cache/apt,sharing=locked \
--mount=type=cache,target=/var/lib/apt,sharing=locked \
apt-get update \
&& apt-get install -y \
--no-install-recommends \
libgl1-mesa-glx=20.3.* \
libglib2.0-0=2.66.* \
libopencv-dev=4.5.*
ENV INVOKEAI_SRC=/opt/invokeai
ENV VIRTUAL_ENV=/opt/venv/invokeai
# Set working directory and env
ARG APPDIR=/usr/src
ARG APPNAME=InvokeAI
WORKDIR ${APPDIR}
ENV PATH ${APPDIR}/${APPNAME}/bin:$PATH
# Keeps Python from generating .pyc files in the container
ENV PYTHONDONTWRITEBYTECODE 1
# Turns off buffering for easier container logging
ENV PYTHONUNBUFFERED 1
# Don't fall back to legacy build system
ENV PIP_USE_PEP517=1
ENV PATH="$VIRTUAL_ENV/bin:$PATH"
ARG TORCH_VERSION=2.0.1
ARG TORCHVISION_VERSION=0.15.2
ARG GPU_DRIVER=cuda
ARG TARGETPLATFORM="linux/amd64"
# unused but available
ARG BUILDPLATFORM
#######################
## build pyproject ##
#######################
FROM python-base AS pyproject-builder
WORKDIR ${INVOKEAI_SRC}
# Install build dependencies
RUN \
--mount=type=cache,target=/var/cache/apt,sharing=locked \
--mount=type=cache,target=/var/lib/apt,sharing=locked \
apt-get update \
&& apt-get install -y \
--no-install-recommends \
build-essential=12.9 \
gcc=4:10.2.* \
python3-dev=3.9.*
# Install pytorch before all other pip packages
# NOTE: there are no pytorch builds for arm64 + cuda, only cpu
# x86_64/CUDA is default
RUN --mount=type=cache,target=/root/.cache/pip \
python3 -m venv ${VIRTUAL_ENV} &&\
if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cpu"; \
elif [ "$GPU_DRIVER" = "rocm" ]; then \
extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/rocm5.4.2"; \
else \
extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cu118"; \
fi &&\
pip install $extra_index_url_arg \
torch==$TORCH_VERSION \
torchvision==$TORCHVISION_VERSION
# Prepare pip for buildkit cache
ARG PIP_CACHE_DIR=/var/cache/buildkit/pip
ENV PIP_CACHE_DIR ${PIP_CACHE_DIR}
RUN mkdir -p ${PIP_CACHE_DIR}
# Install the local package.
# Editable mode helps use the same image for development:
# the local working copy can be bind-mounted into the image
# at path defined by ${INVOKEAI_SRC}
COPY invokeai ./invokeai
COPY pyproject.toml ./
RUN --mount=type=cache,target=/root/.cache/pip \
# xformers + triton fails to install on arm64
if [ "$GPU_DRIVER" = "cuda" ] && [ "$TARGETPLATFORM" = "linux/amd64" ]; then \
pip install -e ".[xformers]"; \
else \
pip install -e "."; \
fi
# Create virtual environment
RUN --mount=type=cache,target=${PIP_CACHE_DIR} \
python3 -m venv "${APPNAME}" \
--upgrade-deps
# #### Build the Web UI ------------------------------------
# Install requirements
COPY --link pyproject.toml .
COPY --link invokeai/version/invokeai_version.py invokeai/version/__init__.py invokeai/version/
ARG PIP_EXTRA_INDEX_URL
ENV PIP_EXTRA_INDEX_URL ${PIP_EXTRA_INDEX_URL}
RUN --mount=type=cache,target=${PIP_CACHE_DIR} \
"${APPNAME}"/bin/pip install .
FROM node:18 AS web-builder
WORKDIR /build
COPY invokeai/frontend/web/ ./
RUN --mount=type=cache,target=/usr/lib/node_modules \
npm install --include dev
RUN --mount=type=cache,target=/usr/lib/node_modules \
yarn vite build
# Install pyproject.toml
COPY --link . .
RUN --mount=type=cache,target=${PIP_CACHE_DIR} \
"${APPNAME}/bin/pip" install .
# Build patchmatch
#### Runtime stage ---------------------------------------
FROM library/ubuntu:22.04 AS runtime
ARG DEBIAN_FRONTEND=noninteractive
ENV PYTHONUNBUFFERED=1
ENV PYTHONDONTWRITEBYTECODE=1
RUN apt update && apt install -y --no-install-recommends \
git \
curl \
vim \
tmux \
ncdu \
iotop \
bzip2 \
gosu \
libglib2.0-0 \
libgl1-mesa-glx \
python3-venv \
python3-pip \
build-essential \
libopencv-dev \
libstdc++-10-dev &&\
apt-get clean && apt-get autoclean
# globally add magic-wormhole
# for ease of transferring data to and from the container
# when running in sandboxed cloud environments; e.g. Runpod etc.
RUN pip install magic-wormhole
ENV INVOKEAI_SRC=/opt/invokeai
ENV VIRTUAL_ENV=/opt/venv/invokeai
ENV INVOKEAI_ROOT=/invokeai
ENV PATH="$VIRTUAL_ENV/bin:$INVOKEAI_SRC:$PATH"
# --link requires buldkit w/ dockerfile syntax 1.4
COPY --link --from=builder ${INVOKEAI_SRC} ${INVOKEAI_SRC}
COPY --link --from=builder ${VIRTUAL_ENV} ${VIRTUAL_ENV}
COPY --link --from=web-builder /build/dist ${INVOKEAI_SRC}/invokeai/frontend/web/dist
# Link amdgpu.ids for ROCm builds
# contributed by https://github.com/Rubonnek
RUN mkdir -p "/opt/amdgpu/share/libdrm" &&\
ln -s "/usr/share/libdrm/amdgpu.ids" "/opt/amdgpu/share/libdrm/amdgpu.ids"
WORKDIR ${INVOKEAI_SRC}
# build patchmatch
RUN cd /usr/lib/$(uname -p)-linux-gnu/pkgconfig/ && ln -sf opencv4.pc opencv.pc
RUN python3 -c "from patchmatch import patch_match"
#####################
## runtime image ##
#####################
FROM python-base AS runtime
# Create unprivileged user and make the local dir
RUN useradd --create-home --shell /bin/bash -u 1000 --comment "container local user" invoke
RUN mkdir -p ${INVOKEAI_ROOT} && chown -R invoke:invoke ${INVOKEAI_ROOT}
# Create a new user
ARG UNAME=appuser
RUN useradd \
--no-log-init \
-m \
-U \
"${UNAME}"
# Create volume directory
ARG VOLUME_DIR=/data
RUN mkdir -p "${VOLUME_DIR}" \
&& chown -hR "${UNAME}:${UNAME}" "${VOLUME_DIR}"
# Setup runtime environment
USER ${UNAME}:${UNAME}
COPY --chown=${UNAME}:${UNAME} --from=pyproject-builder ${APPDIR}/${APPNAME} ${APPNAME}
ENV INVOKEAI_ROOT ${VOLUME_DIR}
ENV TRANSFORMERS_CACHE ${VOLUME_DIR}/.cache
ENV INVOKE_MODEL_RECONFIGURE "--yes --default_only"
EXPOSE 9090
ENTRYPOINT [ "invokeai" ]
CMD [ "--web", "--host", "0.0.0.0", "--port", "9090" ]
VOLUME [ "${VOLUME_DIR}" ]
COPY docker/docker-entrypoint.sh ./
ENTRYPOINT ["/opt/invokeai/docker-entrypoint.sh"]
CMD ["invokeai-web", "--host", "0.0.0.0"]

77
docker/README.md Normal file
View File

@ -0,0 +1,77 @@
# InvokeAI Containerized
All commands are to be run from the `docker` directory: `cd docker`
#### Linux
1. Ensure builkit is enabled in the Docker daemon settings (`/etc/docker/daemon.json`)
2. Install the `docker compose` plugin using your package manager, or follow a [tutorial](https://www.digitalocean.com/community/tutorials/how-to-install-and-use-docker-compose-on-ubuntu-22-04).
- The deprecated `docker-compose` (hyphenated) CLI continues to work for now.
3. Ensure docker daemon is able to access the GPU.
- You may need to install [nvidia-container-toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html)
#### macOS
1. Ensure Docker has at least 16GB RAM
2. Enable VirtioFS for file sharing
3. Enable `docker compose` V2 support
This is done via Docker Desktop preferences
## Quickstart
1. Make a copy of `env.sample` and name it `.env` (`cp env.sample .env` (Mac/Linux) or `copy example.env .env` (Windows)). Make changes as necessary. Set `INVOKEAI_ROOT` to an absolute path to:
a. the desired location of the InvokeAI runtime directory, or
b. an existing, v3.0.0 compatible runtime directory.
1. `docker compose up`
The image will be built automatically if needed.
The runtime directory (holding models and outputs) will be created in the location specified by `INVOKEAI_ROOT`. The default location is `~/invokeai`. The runtime directory will be populated with the base configs and models necessary to start generating.
### Use a GPU
- Linux is *recommended* for GPU support in Docker.
- WSL2 is *required* for Windows.
- only `x86_64` architecture is supported.
The Docker daemon on the system must be already set up to use the GPU. In case of Linux, this involves installing `nvidia-docker-runtime` and configuring the `nvidia` runtime as default. Steps will be different for AMD. Please see Docker documentation for the most up-to-date instructions for using your GPU with Docker.
## Customize
Check the `.env.sample` file. It contains some environment variables for running in Docker. Copy it, name it `.env`, and fill it in with your own values. Next time you run `docker compose up`, your custom values will be used.
You can also set these values in `docker compose.yml` directly, but `.env` will help avoid conflicts when code is updated.
Example (most values are optional):
```
INVOKEAI_ROOT=/Volumes/WorkDrive/invokeai
HUGGINGFACE_TOKEN=the_actual_token
CONTAINER_UID=1000
GPU_DRIVER=cuda
```
## Even Moar Customizing!
See the `docker compose.yaml` file. The `command` instruction can be uncommented and used to run arbitrary startup commands. Some examples below.
### Reconfigure the runtime directory
Can be used to download additional models from the supported model list
In conjunction with `INVOKEAI_ROOT` can be also used to initialize a runtime directory
```
command:
- invokeai-configure
- --yes
```
Or install models:
```
command:
- invokeai-model-install
```

50
docker/build.sh
View File

@ -1,51 +1,11 @@
#!/usr/bin/env bash
set -e
# If you want to build a specific flavor, set the CONTAINER_FLAVOR environment variable
# e.g. CONTAINER_FLAVOR=cpu ./build.sh
# Possible Values are:
# - cpu
# - cuda
# - rocm
# Don't forget to also set it when executing run.sh
# if it is not set, the script will try to detect the flavor by itself.
#
# Doc can be found here:
# https://invoke-ai.github.io/InvokeAI/installation/040_INSTALL_DOCKER/
build_args=""
SCRIPTDIR=$(dirname "${BASH_SOURCE[0]}")
cd "$SCRIPTDIR" || exit 1
[[ -f ".env" ]] && build_args=$(awk '$1 ~ /\=[^$]/ {print "--build-arg " $0 " "}' .env)
source ./env.sh
echo "docker-compose build args:"
echo $build_args
DOCKERFILE=${INVOKE_DOCKERFILE:-./Dockerfile}
# print the settings
echo -e "You are using these values:\n"
echo -e "Dockerfile:\t\t${DOCKERFILE}"
echo -e "index-url:\t\t${PIP_EXTRA_INDEX_URL:-none}"
echo -e "Volumename:\t\t${VOLUMENAME}"
echo -e "Platform:\t\t${PLATFORM}"
echo -e "Container Registry:\t${CONTAINER_REGISTRY}"
echo -e "Container Repository:\t${CONTAINER_REPOSITORY}"
echo -e "Container Tag:\t\t${CONTAINER_TAG}"
echo -e "Container Flavor:\t${CONTAINER_FLAVOR}"
echo -e "Container Image:\t${CONTAINER_IMAGE}\n"
# Create docker volume
if [[ -n "$(docker volume ls -f name="${VOLUMENAME}" -q)" ]]; then
echo -e "Volume already exists\n"
else
echo -n "creating docker volume "
docker volume create "${VOLUMENAME}"
fi
# Build Container
docker build \
--platform="${PLATFORM:-linux/amd64}" \
--tag="${CONTAINER_IMAGE:-invokeai}" \
${CONTAINER_FLAVOR:+--build-arg="CONTAINER_FLAVOR=${CONTAINER_FLAVOR}"} \
${PIP_EXTRA_INDEX_URL:+--build-arg="PIP_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL}"} \
${PIP_PACKAGE:+--build-arg="PIP_PACKAGE=${PIP_PACKAGE}"} \
--file="${DOCKERFILE}" \
..
docker-compose build $build_args

48
docker/docker-compose.yml Normal file
View File

@ -0,0 +1,48 @@
# Copyright (c) 2023 Eugene Brodsky https://github.com/ebr
version: '3.8'
services:
invokeai:
image: "local/invokeai:latest"
# edit below to run on a container runtime other than nvidia-container-runtime.
# not yet tested with rocm/AMD GPUs
# Comment out the "deploy" section to run on CPU only
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: 1
capabilities: [gpu]
build:
context: ..
dockerfile: docker/Dockerfile
# variables without a default will automatically inherit from the host environment
environment:
- INVOKEAI_ROOT
- HF_HOME
# Create a .env file in the same directory as this docker-compose.yml file
# and populate it with environment variables. See .env.sample
env_file:
- .env
ports:
- "${INVOKEAI_PORT:-9090}:9090"
volumes:
- ${INVOKEAI_ROOT:-~/invokeai}:${INVOKEAI_ROOT:-/invokeai}
- ${HF_HOME:-~/.cache/huggingface}:${HF_HOME:-/invokeai/.cache/huggingface}
# - ${INVOKEAI_MODELS_DIR:-${INVOKEAI_ROOT:-/invokeai/models}}
# - ${INVOKEAI_MODELS_CONFIG_PATH:-${INVOKEAI_ROOT:-/invokeai/configs/models.yaml}}
tty: true
stdin_open: true
# # Example of running alternative commands/scripts in the container
# command:
# - bash
# - -c
# - |
# invokeai-model-install --yes --default-only --config_file ${INVOKEAI_ROOT}/config_custom.yaml
# invokeai-nodes-web --host 0.0.0.0

65
docker/docker-entrypoint.sh Executable file
View File

@ -0,0 +1,65 @@
#!/bin/bash
set -e -o pipefail
### Container entrypoint
# Runs the CMD as defined by the Dockerfile or passed to `docker run`
# Can be used to configure the runtime dir
# Bypass by using ENTRYPOINT or `--entrypoint`
### Set INVOKEAI_ROOT pointing to a valid runtime directory
# Otherwise configure the runtime dir first.
### Configure the InvokeAI runtime directory (done by default)):
# docker run --rm -it <this image> --configure
# or skip with --no-configure
### Set the CONTAINER_UID envvar to match your user.
# Ensures files created in the container are owned by you:
# docker run --rm -it -v /some/path:/invokeai -e CONTAINER_UID=$(id -u) <this image>
# Default UID: 1000 chosen due to popularity on Linux systems. Possibly 501 on MacOS.
USER_ID=${CONTAINER_UID:-1000}
USER=invoke
usermod -u ${USER_ID} ${USER} 1>/dev/null
configure() {
# Configure the runtime directory
if [[ -f ${INVOKEAI_ROOT}/invokeai.yaml ]]; then
echo "${INVOKEAI_ROOT}/invokeai.yaml exists. InvokeAI is already configured."
echo "To reconfigure InvokeAI, delete the above file."
echo "======================================================================"
else
mkdir -p ${INVOKEAI_ROOT}
chown --recursive ${USER} ${INVOKEAI_ROOT}
gosu ${USER} invokeai-configure --yes --default_only
fi
}
## Skip attempting to configure.
## Must be passed first, before any other args.
if [[ $1 != "--no-configure" ]]; then
configure
else
shift
fi
### Set the $PUBLIC_KEY env var to enable SSH access.
# We do not install openssh-server in the image by default to avoid bloat.
# but it is useful to have the full SSH server e.g. on Runpod.
# (use SCP to copy files to/from the image, etc)
if [[ -v "PUBLIC_KEY" ]] && [[ ! -d "${HOME}/.ssh" ]]; then
apt-get update
apt-get install -y openssh-server
pushd $HOME
mkdir -p .ssh
echo ${PUBLIC_KEY} > .ssh/authorized_keys
chmod -R 700 .ssh
popd
service ssh start
fi
cd ${INVOKEAI_ROOT}
# Run the CMD as the Container User (not root).
exec gosu ${USER} "$@"

54
docker/env.sh
View File

@ -1,54 +0,0 @@
#!/usr/bin/env bash
# This file is used to set environment variables for the build.sh and run.sh scripts.
# Try to detect the container flavor if no PIP_EXTRA_INDEX_URL got specified
if [[ -z "$PIP_EXTRA_INDEX_URL" ]]; then
# Activate virtual environment if not already activated and exists
if [[ -z $VIRTUAL_ENV ]]; then
[[ -e "$(dirname "${BASH_SOURCE[0]}")/../.venv/bin/activate" ]] \
&& source "$(dirname "${BASH_SOURCE[0]}")/../.venv/bin/activate" \
&& echo "Activated virtual environment: $VIRTUAL_ENV"
fi
# Decide which container flavor to build if not specified
if [[ -z "$CONTAINER_FLAVOR" ]] && python -c "import torch" &>/dev/null; then
# Check for CUDA and ROCm
CUDA_AVAILABLE=$(python -c "import torch;print(torch.cuda.is_available())")
ROCM_AVAILABLE=$(python -c "import torch;print(torch.version.hip is not None)")
if [[ "${CUDA_AVAILABLE}" == "True" ]]; then
CONTAINER_FLAVOR="cuda"
elif [[ "${ROCM_AVAILABLE}" == "True" ]]; then
CONTAINER_FLAVOR="rocm"
else
CONTAINER_FLAVOR="cpu"
fi
fi
# Set PIP_EXTRA_INDEX_URL based on container flavor
if [[ "$CONTAINER_FLAVOR" == "rocm" ]]; then
PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/rocm"
elif [[ "$CONTAINER_FLAVOR" == "cpu" ]]; then
PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu"
# elif [[ -z "$CONTAINER_FLAVOR" || "$CONTAINER_FLAVOR" == "cuda" ]]; then
# PIP_PACKAGE=${PIP_PACKAGE-".[xformers]"}
fi
fi
# Variables shared by build.sh and run.sh
REPOSITORY_NAME="${REPOSITORY_NAME-$(basename "$(git rev-parse --show-toplevel)")}"
REPOSITORY_NAME="${REPOSITORY_NAME,,}"
VOLUMENAME="${VOLUMENAME-"${REPOSITORY_NAME}_data"}"
ARCH="${ARCH-$(uname -m)}"
PLATFORM="${PLATFORM-linux/${ARCH}}"
INVOKEAI_BRANCH="${INVOKEAI_BRANCH-$(git branch --show)}"
CONTAINER_REGISTRY="${CONTAINER_REGISTRY-"ghcr.io"}"
CONTAINER_REPOSITORY="${CONTAINER_REPOSITORY-"$(whoami)/${REPOSITORY_NAME}"}"
CONTAINER_FLAVOR="${CONTAINER_FLAVOR-cuda}"
CONTAINER_TAG="${CONTAINER_TAG-"${INVOKEAI_BRANCH##*/}-${CONTAINER_FLAVOR}"}"
CONTAINER_IMAGE="${CONTAINER_REGISTRY}/${CONTAINER_REPOSITORY}:${CONTAINER_TAG}"
CONTAINER_IMAGE="${CONTAINER_IMAGE,,}"
# enable docker buildkit
export DOCKER_BUILDKIT=1

37
docker/run.sh
View File

@ -1,41 +1,8 @@
#!/usr/bin/env bash
set -e
# How to use: https://invoke-ai.github.io/InvokeAI/installation/040_INSTALL_DOCKER/
SCRIPTDIR=$(dirname "${BASH_SOURCE[0]}")
cd "$SCRIPTDIR" || exit 1
source ./env.sh
# Create outputs directory if it does not exist
[[ -d ./outputs ]] || mkdir ./outputs
echo -e "You are using these values:\n"
echo -e "Volumename:\t${VOLUMENAME}"
echo -e "Invokeai_tag:\t${CONTAINER_IMAGE}"
echo -e "local Models:\t${MODELSPATH:-unset}\n"
docker run \
--interactive \
--tty \
--rm \
--platform="${PLATFORM}" \
--name="${REPOSITORY_NAME}" \
--hostname="${REPOSITORY_NAME}" \
--mount type=volume,volume-driver=local,source="${VOLUMENAME}",target=/data \
--mount type=bind,source="$(pwd)"/outputs/,target=/data/outputs/ \
${MODELSPATH:+--mount="type=bind,source=${MODELSPATH},target=/data/models"} \
${HUGGING_FACE_HUB_TOKEN:+--env="HUGGING_FACE_HUB_TOKEN=${HUGGING_FACE_HUB_TOKEN}"} \
--publish=9090:9090 \
--cap-add=sys_nice \
${GPU_FLAGS:+--gpus="${GPU_FLAGS}"} \
"${CONTAINER_IMAGE}" ${@:+$@}
echo -e "\nCleaning trash folder ..."
for f in outputs/.Trash*; do
if [ -e "$f" ]; then
rm -Rf "$f"
break
fi
done
docker-compose up --build -d
docker-compose logs -f

60
docker/runpod-readme.md Normal file
View File

@ -0,0 +1,60 @@
# InvokeAI - A Stable Diffusion Toolkit
Stable Diffusion distribution by InvokeAI: https://github.com/invoke-ai
The Docker image tracks the `main` branch of the InvokeAI project, which means it includes the latest features, but may contain some bugs.
Your working directory is mounted under the `/workspace` path inside the pod. The models are in `/workspace/invokeai/models`, and outputs are in `/workspace/invokeai/outputs`.
> **Only the /workspace directory will persist between pod restarts!**
> **If you _terminate_ (not just _stop_) the pod, the /workspace will be lost.**
## Quickstart
1. Launch a pod from this template. **It will take about 5-10 minutes to run through the initial setup**. Be patient.
1. Wait for the application to load.
- TIP: you know it's ready when the CPU usage goes idle
- You can also check the logs for a line that says "_Point your browser at..._"
1. Open the Invoke AI web UI: click the `Connect` => `connect over HTTP` button.
1. Generate some art!
## Other things you can do
At any point you may edit the pod configuration and set an arbitrary Docker command. For example, you could run a command to downloads some models using `curl`, or fetch some images and place them into your outputs to continue a working session.
If you need to run *multiple commands*, define them in the Docker Command field like this:
`bash -c "cd ${INVOKEAI_ROOT}/outputs; wormhole receive 2-foo-bar; invoke.py --web --host 0.0.0.0"`
### Copying your data in and out of the pod
This image includes a couple of handy tools to help you get the data into the pod (such as your custom models or embeddings), and out of the pod (such as downloading your outputs). Here are your options for getting your data in and out of the pod:
- **SSH server**:
1. Make sure to create and set your Public Key in the RunPod settings (follow the official instructions)
1. Add an exposed port 22 (TCP) in the pod settings!
1. When your pod restarts, you will see a new entry in the `Connect` dialog. Use this SSH server to `scp` or `sftp` your files as necessary, or SSH into the pod using the fully fledged SSH server.
- [**Magic Wormhole**](https://magic-wormhole.readthedocs.io/en/latest/welcome.html):
1. On your computer, `pip install magic-wormhole` (see above instructions for details)
1. Connect to the command line **using the "light" SSH client** or the browser-based console. _Currently there's a bug where `wormhole` isn't available when connected to "full" SSH server, as described above_.
1. `wormhole send /workspace/invokeai/outputs` will send the entire `outputs` directory. You can also send individual files.
1. Once packaged, you will see a `wormhole receive <123-some-words>` command. Copy it
1. Paste this command into the terminal on your local machine to securely download the payload.
1. It works the same in reverse: you can `wormhole send` some models from your computer to the pod. Again, save your files somewhere in `/workspace` or they will be lost when the pod is stopped.
- **RunPod's Cloud Sync feature** may be used to sync the persistent volume to cloud storage. You could, for example, copy the entire `/workspace` to S3, add some custom models to it, and copy it back from S3 when launching new pod configurations. Follow the Cloud Sync instructions.
### Disable the NSFW checker
The NSFW checker is enabled by default. To disable it, edit the pod configuration and set the following command:
```
invoke --web --host 0.0.0.0 --no-nsfw_checker
```
---
Template ©2023 Eugene Brodsky [ebr](https://github.com/ebr)

238
docs/CHANGELOG.md
View File

@ -4,6 +4,236 @@ title: Changelog
# :octicons-log-16: **Changelog**
## v2.3.5 <small>(22 May 2023)</small>
This release (along with the post1 and post2 follow-on releases) expands support for additional LoRA and LyCORIS models, upgrades diffusers versions, and fixes a few bugs.
### LoRA and LyCORIS Support Improvement
A number of LoRA/LyCORIS fine-tune files (those which alter the text encoder as well as the unet model) were not having the desired effect in InvokeAI. This bug has now been fixed. Full documentation of LoRA support is available at InvokeAI LoRA Support.
Previously, InvokeAI did not distinguish between LoRA/LyCORIS models based on Stable Diffusion v1.5 vs those based on v2.0 and 2.1, leading to a crash when an incompatible model was loaded. This has now been fixed. In addition, the web pulldown menus for LoRA and Textual Inversion selection have been enhanced to show only those files that are compatible with the currently-selected Stable Diffusion model.
Support for the newer LoKR LyCORIS files has been added.
### Library Updates and Speed/Reproducibility Advancements
The major enhancement in this version is that NVIDIA users no longer need to decide between speed and reproducibility. Previously, if you activated the Xformers library, you would see improvements in speed and memory usage, but multiple images generated with the same seed and other parameters would be slightly different from each other. This is no longer the case. Relative to 2.3.5 you will see improved performance when running without Xformers, and even better performance when Xformers is activated. In both cases, images generated with the same settings will be identical.
Here are the new library versions:
Library Version
Torch 2.0.0
Diffusers 0.16.1
Xformers 0.0.19
Compel 1.1.5
Other Improvements
### Performance Improvements
When a model is loaded for the first time, InvokeAI calculates its checksum for incorporation into the PNG metadata. This process could take up to a minute on network-mounted disks and WSL mounts. This release noticeably speeds up the process.
### Bug Fixes
The "import models from directory" and "import from URL" functionality in the console-based model installer has now been fixed.
When running the WebUI, we have reduced the number of times that InvokeAI reaches out to HuggingFace to fetch the list of embeddable Textual Inversion models. We have also caught and fixed a problem with the updater not correctly detecting when another instance of the updater is running
## v2.3.4 <small>(7 April 2023)</small>
What's New in 2.3.4
This features release adds support for LoRA (Low-Rank Adaptation) and LyCORIS (Lora beYond Conventional) models, as well as some minor bug fixes.
### LoRA and LyCORIS Support
LoRA files contain fine-tuning weights that enable particular styles, subjects or concepts to be applied to generated images. LyCORIS files are an extended variant of LoRA. InvokeAI supports the most common LoRA/LyCORIS format, which ends in the suffix .safetensors. You will find numerous LoRA and LyCORIS models for download at Civitai, and a small but growing number at Hugging Face. Full documentation of LoRA support is available at InvokeAI LoRA Support.( Pre-release note: this page will only be available after release)
To use LoRA/LyCORIS models in InvokeAI:
Download the .safetensors files of your choice and place in /path/to/invokeai/loras. This directory was not present in earlier version of InvokeAI but will be created for you the first time you run the command-line or web client. You can also create the directory manually.
Add withLora(lora-file,weight) to your prompts. The weight is optional and will default to 1.0. A few examples, assuming that a LoRA file named loras/sushi.safetensors is present:
family sitting at dinner table eating sushi withLora(sushi,0.9)
family sitting at dinner table eating sushi withLora(sushi, 0.75)
family sitting at dinner table eating sushi withLora(sushi)
Multiple withLora() prompt fragments are allowed. The weight can be arbitrarily large, but the useful range is roughly 0.5 to 1.0. Higher weights make the LoRA's influence stronger. Negative weights are also allowed, which can lead to some interesting effects.
Generate as you usually would! If you find that the image is too "crisp" try reducing the overall CFG value or reducing individual LoRA weights. As is the case with all fine-tunes, you'll get the best results when running the LoRA on top of the model similar to, or identical with, the one that was used during the LoRA's training. Don't try to load a SD 1.x-trained LoRA into a SD 2.x model, and vice versa. This will trigger a non-fatal error message and generation will not proceed.
You can change the location of the loras directory by passing the --lora_directory option to `invokeai.
### New WebUI LoRA and Textual Inversion Buttons
This version adds two new web interface buttons for inserting LoRA and Textual Inversion triggers into the prompt as shown in the screenshot below.
Clicking on one or the other of the buttons will bring up a menu of available LoRA/LyCORIS or Textual Inversion trigger terms. Select a menu item to insert the properly-formatted withLora() or <textual-inversion> prompt fragment into the positive prompt. The number in parentheses indicates the number of trigger terms currently in the prompt. You may click the button again and deselect the LoRA or trigger to remove it from the prompt, or simply edit the prompt directly.
Currently terms are inserted into the positive prompt textbox only. However, some textual inversion embeddings are designed to be used with negative prompts. To move a textual inversion trigger into the negative prompt, simply cut and paste it.
By default the Textual Inversion menu only shows locally installed models found at startup time in /path/to/invokeai/embeddings. However, InvokeAI has the ability to dynamically download and install additional Textual Inversion embeddings from the HuggingFace Concepts Library. You may choose to display the most popular of these (with five or more likes) in the Textual Inversion menu by going to Settings and turning on "Show Textual Inversions from HF Concepts Library." When this option is activated, the locally-installed TI embeddings will be shown first, followed by uninstalled terms from Hugging Face. See The Hugging Face Concepts Library and Importing Textual Inversion files for more information.
### Minor features and fixes
This release changes model switching behavior so that the command-line and Web UIs save the last model used and restore it the next time they are launched. It also improves the behavior of the installer so that the pip utility is kept up to date.
### Known Bugs in 2.3.4
These are known bugs in the release.
The Ancestral DPMSolverMultistepScheduler (k_dpmpp_2a) sampler is not yet implemented for diffusers models and will disappear from the WebUI Sampler menu when a diffusers model is selected.
Windows Defender will sometimes raise Trojan or backdoor alerts for the codeformer.pth face restoration model, as well as the CIDAS/clipseg and runwayml/stable-diffusion-v1.5 models. These are false positives and can be safely ignored. InvokeAI performs a malware scan on all models as they are loaded. For additional security, you should use safetensors models whenever they are available.
## v2.3.3 <small>(28 March 2023)</small>
This is a bugfix and minor feature release.
### Bugfixes
Since version 2.3.2 the following bugs have been fixed:
Bugs
When using legacy checkpoints with an external VAE, the VAE file is now scanned for malware prior to loading. Previously only the main model weights file was scanned.
Textual inversion will select an appropriate batchsize based on whether xformers is active, and will default to xformers enabled if the library is detected.
The batch script log file names have been fixed to be compatible with Windows.
Occasional corruption of the .next_prefix file (which stores the next output file name in sequence) on Windows systems is now detected and corrected.
Support loading of legacy config files that have no personalization (textual inversion) section.
An infinite loop when opening the developer's console from within the invoke.sh script has been corrected.
Documentation fixes, including a recipe for detecting and fixing problems with the AMD GPU ROCm driver.
Enhancements
It is now possible to load and run several community-contributed SD-2.0 based models, including the often-requested "Illuminati" model.
The "NegativePrompts" embedding file, and others like it, can now be loaded by placing it in the InvokeAI embeddings directory.
If no --model is specified at launch time, InvokeAI will remember the last model used and restore it the next time it is launched.
On Linux systems, the invoke.sh launcher now uses a prettier console-based interface. To take advantage of it, install the dialog package using your package manager (e.g. sudo apt install dialog).
When loading legacy models (safetensors/ckpt) you can specify a custom config file and/or a VAE by placing like-named files in the same directory as the model following this example:
my-favorite-model.ckpt
my-favorite-model.yaml
my-favorite-model.vae.pt # or my-favorite-model.vae.safetensors
### Known Bugs in 2.3.3
These are known bugs in the release.
The Ancestral DPMSolverMultistepScheduler (k_dpmpp_2a) sampler is not yet implemented for diffusers models and will disappear from the WebUI Sampler menu when a diffusers model is selected.
Windows Defender will sometimes raise Trojan or backdoor alerts for the codeformer.pth face restoration model, as well as the CIDAS/clipseg and runwayml/stable-diffusion-v1.5 models. These are false positives and can be safely ignored. InvokeAI performs a malware scan on all models as they are loaded. For additional security, you should use safetensors models whenever they are available.
## v2.3.2 <small>(11 March 2023)</small>
This is a bugfix and minor feature release.
### Bugfixes
Since version 2.3.1 the following bugs have been fixed:
Black images appearing for potential NSFW images when generating with legacy checkpoint models and both --no-nsfw_checker and --ckpt_convert turned on.
Black images appearing when generating from models fine-tuned on Stable-Diffusion-2-1-base. When importing V2-derived models, you may be asked to select whether the model was derived from a "base" model (512 pixels) or the 768-pixel SD-2.1 model.
The "Use All" button was not restoring the Hi-Res Fix setting on the WebUI
When using the model installer console app, models failed to import correctly when importing from directories with spaces in their names. A similar issue with the output directory was also fixed.
Crashes that occurred during model merging.
Restore previous naming of Stable Diffusion base and 768 models.
Upgraded to latest versions of diffusers, transformers, safetensors and accelerate libraries upstream. We hope that this will fix the assertion NDArray > 2**32 issue that MacOS users have had when generating images larger than 768x768 pixels. Please report back.
As part of the upgrade to diffusers, the location of the diffusers-based models has changed from models/diffusers to models/hub. When you launch InvokeAI for the first time, it will prompt you to OK a one-time move. This should be quick and harmless, but if you have modified your models/diffusers directory in some way, for example using symlinks, you may wish to cancel the migration and make appropriate adjustments.
New "Invokeai-batch" script
### Invoke AI Batch
2.3.2 introduces a new command-line only script called invokeai-batch that can be used to generate hundreds of images from prompts and settings that vary systematically. This can be used to try the same prompt across multiple combinations of models, steps, CFG settings and so forth. It also allows you to template prompts and generate a combinatorial list like:
a shack in the mountains, photograph
a shack in the mountains, watercolor
a shack in the mountains, oil painting
a chalet in the mountains, photograph
a chalet in the mountains, watercolor
a chalet in the mountains, oil painting
a shack in the desert, photograph
...
If you have a system with multiple GPUs, or a single GPU with lots of VRAM, you can parallelize generation across the combinatorial set, reducing wait times and using your system's resources efficiently (make sure you have good GPU cooling).
To try invokeai-batch out. Launch the "developer's console" using the invoke launcher script, or activate the invokeai virtual environment manually. From the console, give the command invokeai-batch --help in order to learn how the script works and create your first template file for dynamic prompt generation.
### Known Bugs in 2.3.2
These are known bugs in the release.
The Ancestral DPMSolverMultistepScheduler (k_dpmpp_2a) sampler is not yet implemented for diffusers models and will disappear from the WebUI Sampler menu when a diffusers model is selected.
Windows Defender will sometimes raise a Trojan alert for the codeformer.pth face restoration model. As far as we have been able to determine, this is a false positive and can be safely whitelisted.
## v2.3.1 <small>(22 February 2023)</small>
This is primarily a bugfix release, but it does provide several new features that will improve the user experience.
### Enhanced support for model management
InvokeAI now makes it convenient to add, remove and modify models. You can individually import models that are stored on your local system, scan an entire folder and its subfolders for models and import them automatically, and even directly import models from the internet by providing their download URLs. You also have the option of designating a local folder to scan for new models each time InvokeAI is restarted.
There are three ways of accessing the model management features:
From the WebUI, click on the cube to the right of the model selection menu. This will bring up a form that allows you to import models individually from your local disk or scan a directory for models to import.
Using the Model Installer App
Choose option (5) download and install models from the invoke launcher script to start a new console-based application for model management. You can use this to select from a curated set of starter models, or import checkpoint, safetensors, and diffusers models from a local disk or the internet. The example below shows importing two checkpoint URLs from popular SD sites and a HuggingFace diffusers model using its Repository ID. It also shows how to designate a folder to be scanned at startup time for new models to import.
Command-line users can start this app using the command invokeai-model-install.
Using the Command Line Client (CLI)
The !install_model and !convert_model commands have been enhanced to allow entering of URLs and local directories to scan and import. The first command installs .ckpt and .safetensors files as-is. The second one converts them into the faster diffusers format before installation.
Internally InvokeAI is able to probe the contents of a .ckpt or .safetensors file to distinguish among v1.x, v2.x and inpainting models. This means that you do not need to include "inpaint" in your model names to use an inpainting model. Note that Stable Diffusion v2.x models will be autoconverted into a diffusers model the first time you use it.
Please see INSTALLING MODELS for more information on model management.
### An Improved Installer Experience
The installer now launches a console-based UI for setting and changing commonly-used startup options:
After selecting the desired options, the installer installs several support models needed by InvokeAI's face reconstruction and upscaling features and then launches the interface for selecting and installing models shown earlier. At any time, you can edit the startup options by launching invoke.sh/invoke.bat and entering option (6) change InvokeAI startup options
Command-line users can launch the new configure app using invokeai-configure.
This release also comes with a renewed updater. To do an update without going through a whole reinstallation, launch invoke.sh or invoke.bat and choose option (9) update InvokeAI . This will bring you to a screen that prompts you to update to the latest released version, to the most current development version, or any released or unreleased version you choose by selecting the tag or branch of the desired version.
Command-line users can run this interface by typing invokeai-configure
### Image Symmetry Options
There are now features to generate horizontal and vertical symmetry during generation. The way these work is to wait until a selected step in the generation process and then to turn on a mirror image effect. In addition to generating some cool images, you can also use this to make side-by-side comparisons of how an image will look with more or fewer steps. Access this option from the WebUI by selecting Symmetry from the image generation settings, or within the CLI by using the options --h_symmetry_time_pct and --v_symmetry_time_pct (these can be abbreviated to --h_sym and --v_sym like all other options).
### A New Unified Canvas Look
This release introduces a beta version of the WebUI Unified Canvas. To try it out, open up the settings dialogue in the WebUI (gear icon) and select Use Canvas Beta Layout:
Refresh the screen and go to to Unified Canvas (left side of screen, third icon from the top). The new layout is designed to provide more space to work in and to keep the image controls close to the image itself:
Model conversion and merging within the WebUI
The WebUI now has an intuitive interface for model merging, as well as for permanent conversion of models from legacy .ckpt/.safetensors formats into diffusers format. These options are also available directly from the invoke.sh/invoke.bat scripts.
An easier way to contribute translations to the WebUI
We have migrated our translation efforts to Weblate, a FOSS translation product. Maintaining the growing project's translations is now far simpler for the maintainers and community. Please review our brief translation guide for more information on how to contribute.
Numerous internal bugfixes and performance issues
### Bug Fixes
This releases quashes multiple bugs that were reported in 2.3.0. Major internal changes include upgrading to diffusers 0.13.0, and using the compel library for prompt parsing. See Detailed Change Log for a detailed list of bugs caught and squished.
Summary of InvokeAI command line scripts (all accessible via the launcher menu)
Command Description
invokeai Command line interface
invokeai --web Web interface
invokeai-model-install Model installer with console forms-based front end
invokeai-ti --gui Textual inversion, with a console forms-based front end
invokeai-merge --gui Model merging, with a console forms-based front end
invokeai-configure Startup configuration; can also be used to reinstall support models
invokeai-update InvokeAI software updater
### Known Bugs in 2.3.1
These are known bugs in the release.
MacOS users generating 768x768 pixel images or greater using diffusers models may experience a hard crash with assertion NDArray > 2**32 This appears to be an issu...
## v2.3.0 <small>(15 January 2023)</small>
**Transition to diffusers
@ -264,7 +494,7 @@ sections describe what's new for InvokeAI.
[Manual Installation](installation/020_INSTALL_MANUAL.md).
- The ability to save frequently-used startup options (model to load, steps,
sampler, etc) in a `.invokeai` file. See
[Client](features/CLI.md)
[Client](deprecated/CLI.md)
- Support for AMD GPU cards (non-CUDA) on Linux machines.
- Multiple bugs and edge cases squashed.
@ -387,7 +617,7 @@ sections describe what's new for InvokeAI.
- `dream.py` script renamed `invoke.py`. A `dream.py` script wrapper remains for
backward compatibility.
- Completely new WebGUI - launch with `python3 scripts/invoke.py --web`
- Support for [inpainting](features/INPAINTING.md) and
- Support for [inpainting](deprecated/INPAINTING.md) and
[outpainting](features/OUTPAINTING.md)
- img2img runs on all k\* samplers
- Support for
@ -399,7 +629,7 @@ sections describe what's new for InvokeAI.
using facial reconstruction, ESRGAN upscaling, outcropping (similar to DALL-E
infinite canvas), and "embiggen" upscaling. See the `!fix` command.
- New `--hires` option on `invoke>` line allows
[larger images to be created without duplicating elements](features/CLI.md#this-is-an-example-of-txt2img),
[larger images to be created without duplicating elements](deprecated/CLI.md#this-is-an-example-of-txt2img),
at the cost of some performance.
- New `--perlin` and `--threshold` options allow you to add and control
variation during image generation (see
@ -408,7 +638,7 @@ sections describe what's new for InvokeAI.
of images and tweaking of previous settings.
- Command-line completion in `invoke.py` now works on Windows, Linux and Mac
platforms.
- Improved [command-line completion behavior](features/CLI.md) New commands
- Improved [command-line completion behavior](deprecated/CLI.md) New commands
added:
- List command-line history with `!history`
- Search command-line history with `!search`

BIN
docs/assets/contributing/resize_invocation.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 7.1 KiB

BIN
docs/assets/contributing/resize_node_editor.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 17 KiB

BIN
docs/assets/features/restoration-montage.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 4.0 MiB

BIN
docs/assets/features/upscale-dialog.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 310 KiB

BIN
docs/assets/features/upscaling-montage.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 8.3 MiB

54
docs/contributing/CONTRIBUTING.md Normal file
View File

@ -0,0 +1,54 @@
## Welcome to Invoke AI
We're thrilled to have you here and we're excited for you to contribute.
Invoke AI originated as a project built by the community, and that vision carries forward today as we aim to build the best pro-grade tools available. We work together to incorporate the latest in AI/ML research, making these tools available in over 20 languages to artists and creatives around the world as part of our fully permissive OSS project designed for individual users to self-host and use.
Here are some guidelines to help you get started:
### Technical Prerequisites
Front-end: You'll need a working knowledge of React and TypeScript.
Back-end: Depending on the scope of your contribution, you may need to know SQLite, FastAPI, Python, and Socketio. Also, a good majority of the backend logic involved in processing images is built in a modular way using a concept called "Nodes", which are isolated functions that carry out individual, discrete operations. This design allows for easy contributions of novel pipelines and capabilities.
### How to Submit Contributions
To start contributing, please follow these steps:
1. Familiarize yourself with our roadmap and open projects to see where your skills and interests align. These documents can serve as a source of inspiration.
2. Open a Pull Request (PR) with a clear description of the feature you're adding or the problem you're solving. Make sure your contribution aligns with the project's vision.
3. Adhere to general best practices. This includes assuming interoperability with other nodes, keeping the scope of your functions as small as possible, and organizing your code according to our architecture documents.
### Types of Contributions We're Looking For
We welcome all contributions that improve the project. Right now, we're especially looking for:
1. Quality of life (QOL) enhancements on the front-end.
2. New backend capabilities added through nodes.
3. Incorporating additional optimizations from the broader open-source software community.
### Communication and Decision-making Process
Project maintainers and code owners review PRs to ensure they align with the project's goals. They may provide design or architectural guidance, suggestions on user experience, or provide more significant feedback on the contribution itself. Expect to receive feedback on your submissions, and don't hesitate to ask questions or propose changes.
For more robust discussions, or if you're planning to add capabilities not currently listed on our roadmap, please reach out to us on our Discord server. That way, we can ensure your proposed contribution aligns with the project's direction before you start writing code.
### Code of Conduct and Contribution Expectations
We want everyone in our community to have a positive experience. To facilitate this, we've established a code of conduct and a statement of values that we expect all contributors to adhere to. Please take a moment to review these documents—they're essential to maintaining a respectful and inclusive environment.
By making a contribution to this project, you certify that:
1. The contribution was created in whole or in part by you and you have the right to submit it under the open-source license indicated in this project’s GitHub repository; or
2. The contribution is based upon previous work that, to the best of your knowledge, is covered under an appropriate open-source license and you have the right under that license to submit that work with modifications, whether created in whole or in part by you, under the same open-source license (unless you are permitted to submit under a different license); or
3. The contribution was provided directly to you by some other person who certified (1) or (2) and you have not modified it; or
4. You understand and agree that this project and the contribution are public and that a record of the contribution (including all personal information you submit with it, including your sign-off) is maintained indefinitely and may be redistributed consistent with this project or the open-source license(s) involved.
This disclaimer is not a license and does not grant any rights or permissions. You must obtain necessary permissions and licenses, including from third parties, before contributing to this project.
This disclaimer is provided "as is" without warranty of any kind, whether expressed or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, or non-infringement. In no event shall the authors or copyright holders be liable for any claim, damages, or other liability, whether in an action of contract, tort, or otherwise, arising from, out of, or in connection with the contribution or the use or other dealings in the contribution.
---
Remember, your contributions help make this project great. We're excited to see what you'll bring to our community!

519
docs/contributing/INVOCATIONS.md
View File

@ -1,8 +1,521 @@
# Invocations
Invocations represent a single operation, its inputs, and its outputs. These
operations and their outputs can be chained together to generate and modify
images.
Features in InvokeAI are added in the form of modular node-like systems called
**Invocations**.
An Invocation is simply a single operation that takes in some inputs and gives
out some outputs. We can then chain multiple Invocations together to create more
complex functionality.
## Invocations Directory
InvokeAI Invocations can be found in the `invokeai/app/invocations` directory.
You can add your new functionality to one of the existing Invocations in this
directory or create a new file in this directory as per your needs.
**Note:** _All Invocations must be inside this directory for InvokeAI to
recognize them as valid Invocations._
## Creating A New Invocation
In order to understand the process of creating a new Invocation, let us actually
create one.
In our example, let us create an Invocation that will take in an image, resize
it and output the resized image.
The first set of things we need to do when creating a new Invocation are -
- Create a new class that derives from a predefined parent class called
`BaseInvocation`.
- The name of every Invocation must end with the word `Invocation` in order for
it to be recognized as an Invocation.
- Every Invocation must have a `docstring` that describes what this Invocation
does.
- Every Invocation must have a unique `type` field defined which becomes its
indentifier.
- Invocations are strictly typed. We make use of the native
[typing](https://docs.python.org/3/library/typing.html) library and the
installed [pydantic](https://pydantic-docs.helpmanual.io/) library for
validation.
So let us do that.
```python
from typing import Literal
from .baseinvocation import BaseInvocation
class ResizeInvocation(BaseInvocation):
'''Resizes an image'''
type: Literal['resize'] = 'resize'
```
That's great.
Now we have setup the base of our new Invocation. Let us think about what inputs
our Invocation takes.
- We need an `image` that we are going to resize.
- We will need new `width` and `height` values to which we need to resize the
image to.
### **Inputs**
Every Invocation input is a pydantic `Field` and like everything else should be
strictly typed and defined.
So let us create these inputs for our Invocation. First up, the `image` input we
need. Generally, we can use standard variable types in Python but InvokeAI
already has a custom `ImageField` type that handles all the stuff that is needed
for image inputs.
But what is this `ImageField` ..? It is a special class type specifically
written to handle how images are dealt with in InvokeAI. We will cover how to
create your own custom field types later in this guide. For now, let's go ahead
and use it.
```python
from typing import Literal, Union
from pydantic import Field
from .baseinvocation import BaseInvocation
from ..models.image import ImageField
class ResizeInvocation(BaseInvocation):
'''Resizes an image'''
type: Literal['resize'] = 'resize'
# Inputs
image: Union[ImageField, None] = Field(description="The input image", default=None)
```
Let us break down our input code.
```python
image: Union[ImageField, None] = Field(description="The input image", default=None)
```
| Part | Value | Description |
| --------- | ---------------------------------------------------- | -------------------------------------------------------------------------------------------------- |
| Name | `image` | The variable that will hold our image |
| Type Hint | `Union[ImageField, None]` | The types for our field. Indicates that the image can either be an `ImageField` type or `None` |
| Field | `Field(description="The input image", default=None)` | The image variable is a field which needs a description and a default value that we set to `None`. |
Great. Now let us create our other inputs for `width` and `height`
```python
from typing import Literal, Union
from pydantic import Field
from .baseinvocation import BaseInvocation
from ..models.image import ImageField
class ResizeInvocation(BaseInvocation):
'''Resizes an image'''
type: Literal['resize'] = 'resize'
# Inputs
image: Union[ImageField, None] = Field(description="The input image", default=None)
width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
```
As you might have noticed, we added two new parameters to the field type for
`width` and `height` called `gt` and `le`. These basically stand for _greater
than or equal to_ and _less than or equal to_. There are various other param
types for field that you can find on the **pydantic** documentation.
**Note:** _Any time it is possible to define constraints for our field, we
should do it so the frontend has more information on how to parse this field._
Perfect. We now have our inputs. Let us do something with these.
### **Invoke Function**
The `invoke` function is where all the magic happens. This function provides you
the `context` parameter that is of the type `InvocationContext` which will give
you access to the current context of the generation and all the other services
that are provided by it by InvokeAI.
Let us create this function first.
```python
from typing import Literal, Union
from pydantic import Field
from .baseinvocation import BaseInvocation, InvocationContext
from ..models.image import ImageField
class ResizeInvocation(BaseInvocation):
'''Resizes an image'''
type: Literal['resize'] = 'resize'
# Inputs
image: Union[ImageField, None] = Field(description="The input image", default=None)
width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
def invoke(self, context: InvocationContext):
pass
```
### **Outputs**
The output of our Invocation will be whatever is returned by this `invoke`
function. Like with our inputs, we need to strongly type and define our outputs
too.
What is our output going to be? Another image. Normally you'd have to create a
type for this but InvokeAI already offers you an `ImageOutput` type that handles
all the necessary info related to image outputs. So let us use that.
We will cover how to create your own output types later in this guide.
```python
from typing import Literal, Union
from pydantic import Field
from .baseinvocation import BaseInvocation, InvocationContext
from ..models.image import ImageField
from .image import ImageOutput
class ResizeInvocation(BaseInvocation):
'''Resizes an image'''
type: Literal['resize'] = 'resize'
# Inputs
image: Union[ImageField, None] = Field(description="The input image", default=None)
width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
def invoke(self, context: InvocationContext) -> ImageOutput:
pass
```
Perfect. Now that we have our Invocation setup, let us do what we want to do.
- We will first load the image. Generally we do this using the `PIL` library but
we can use one of the services provided by InvokeAI to load the image.
- We will resize the image using `PIL` to our input data.
- We will output this image in the format we set above.
So let's do that.
```python
from typing import Literal, Union
from pydantic import Field
from .baseinvocation import BaseInvocation, InvocationContext
from ..models.image import ImageField, ResourceOrigin, ImageCategory
from .image import ImageOutput
class ResizeInvocation(BaseInvocation):
'''Resizes an image'''
type: Literal['resize'] = 'resize'
# Inputs
image: Union[ImageField, None] = Field(description="The input image", default=None)
width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
def invoke(self, context: InvocationContext) -> ImageOutput:
# Load the image using InvokeAI's predefined Image Service.
image = context.services.images.get_pil_image(self.image.image_origin, self.image.image_name)
# Resizing the image
# Because we used the above service, we already have a PIL image. So we can simply resize.
resized_image = image.resize((self.width, self.height))
# Preparing the image for output using InvokeAI's predefined Image Service.
output_image = context.services.images.create(
image=resized_image,
image_origin=ResourceOrigin.INTERNAL,
image_category=ImageCategory.GENERAL,
node_id=self.id,
session_id=context.graph_execution_state_id,
is_intermediate=self.is_intermediate,
)
# Returning the Image
return ImageOutput(
image=ImageField(
image_name=output_image.image_name,
image_origin=output_image.image_origin,
),
width=output_image.width,
height=output_image.height,
)
```
**Note:** Do not be overwhelmed by the `ImageOutput` process. InvokeAI has a
certain way that the images need to be dispatched in order to be stored and read
correctly. In 99% of the cases when dealing with an image output, you can simply
copy-paste the template above.
That's it. You made your own **Resize Invocation**.
## Result
Once you make your Invocation correctly, the rest of the process is fully
automated for you.
When you launch InvokeAI, you can go to `http://localhost:9090/docs` and see
your new Invocation show up there with all the relevant info.
![resize invocation](../assets/contributing/resize_invocation.png)
When you launch the frontend UI, you can go to the Node Editor tab and find your
new Invocation ready to be used.
![resize node editor](../assets/contributing/resize_node_editor.png)
# Advanced
## Custom Input Fields
Now that you know how to create your own Invocations, let us dive into slightly
more advanced topics.
While creating your own Invocations, you might run into a scenario where the
existing input types in InvokeAI do not meet your requirements. In such cases,
you can create your own input types.
Let us create one as an example. Let us say we want to create a color input
field that represents a color code. But before we start on that here are some
general good practices to keep in mind.
**Good Practices**
- There is no naming convention for input fields but we highly recommend that
you name it something appropriate like `ColorField`.
- It is not mandatory but it is heavily recommended to add a relevant
`docstring` to describe your input field.
- Keep your field in the same file as the Invocation that it is made for or in
another file where it is relevant.
All input types a class that derive from the `BaseModel` type from `pydantic`.
So let's create one.
```python
from pydantic import BaseModel
class ColorField(BaseModel):
'''A field that holds the rgba values of a color'''
pass
```
Perfect. Now let us create our custom inputs for our field. This is exactly
similar how you created input fields for your Invocation. All the same rules
apply. Let us create four fields representing the _red(r)_, _blue(b)_,
_green(g)_ and _alpha(a)_ channel of the color.
```python
class ColorField(BaseModel):
'''A field that holds the rgba values of a color'''
r: int = Field(ge=0, le=255, description="The red channel")
g: int = Field(ge=0, le=255, description="The green channel")
b: int = Field(ge=0, le=255, description="The blue channel")
a: int = Field(ge=0, le=255, description="The alpha channel")
```
That's it. We now have a new input field type that we can use in our Invocations
like this.
```python
color: ColorField = Field(default=ColorField(r=0, g=0, b=0, a=0), description='Background color of an image')
```
**Extra Config**
All input fields also take an additional `Config` class that you can use to do
various advanced things like setting required parameters and etc.
Let us do that for our _ColorField_ and enforce all the values because we did
not define any defaults for our fields.
```python
class ColorField(BaseModel):
'''A field that holds the rgba values of a color'''
r: int = Field(ge=0, le=255, description="The red channel")
g: int = Field(ge=0, le=255, description="The green channel")
b: int = Field(ge=0, le=255, description="The blue channel")
a: int = Field(ge=0, le=255, description="The alpha channel")
class Config:
schema_extra = {"required": ["r", "g", "b", "a"]}
```
Now it becomes mandatory for the user to supply all the values required by our
input field.
We will discuss the `Config` class in extra detail later in this guide and how
you can use it to make your Invocations more robust.
## Custom Output Types
Like with custom inputs, sometimes you might find yourself needing custom
outputs that InvokeAI does not provide. We can easily set one up.
Now that you are familiar with Invocations and Inputs, let us use that knowledge
to put together a custom output type for an Invocation that returns _width_,
_height_ and _background_color_ that we need to create a blank image.
- A custom output type is a class that derives from the parent class of
`BaseInvocationOutput`.
- It is not mandatory but we recommend using names ending with `Output` for
output types. So we'll call our class `BlankImageOutput`
- It is not mandatory but we highly recommend adding a `docstring` to describe
what your output type is for.
- Like Invocations, each output type should have a `type` variable that is
**unique**
Now that we know the basic rules for creating a new output type, let us go ahead
and make it.
```python
from typing import Literal
from pydantic import Field
from .baseinvocation import BaseInvocationOutput
class BlankImageOutput(BaseInvocationOutput):
'''Base output type for creating a blank image'''
type: Literal['blank_image_output'] = 'blank_image_output'
# Inputs
width: int = Field(description='Width of blank image')
height: int = Field(description='Height of blank image')
bg_color: ColorField = Field(description='Background color of blank image')
class Config:
schema_extra = {"required": ["type", "width", "height", "bg_color"]}
```
All set. We now have an output type that requires what we need to create a
blank_image. And if you noticed it, we even used the `Config` class to ensure
the fields are required.
## Custom Configuration
As you might have noticed when making inputs and outputs, we used a class called
`Config` from _pydantic_ to further customize them. Because our inputs and
outputs essentially inherit from _pydantic_'s `BaseModel` class, all
[configuration options](https://docs.pydantic.dev/latest/usage/schema/#schema-customization)
that are valid for _pydantic_ classes are also valid for our inputs and outputs.
You can do the same for your Invocations too but InvokeAI makes our life a
little bit easier on that end.
InvokeAI provides a custom configuration class called `InvocationConfig`
particularly for configuring Invocations. This is exactly the same as the raw
`Config` class from _pydantic_ with some extra stuff on top to help faciliate
parsing of the scheme in the frontend UI.
At the current moment, tihs `InvocationConfig` class is further improved with
the following features related the `ui`.
| Config Option | Field Type | Example |
| ------------- | ------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------- |
| type_hints | `Dict[str, Literal["integer", "float", "boolean", "string", "enum", "image", "latents", "model", "control"]]` | `type_hint: "model"` provides type hints related to the model like displaying a list of available models |
| tags | `List[str]` | `tags: ['resize', 'image']` will classify your invocation under the tags of resize and image. |
| title | `str` | `title: 'Resize Image` will rename your to this custom title rather than infer from the name of the Invocation class. |
So let us update your `ResizeInvocation` with some extra configuration and see
how that works.
```python
from typing import Literal, Union
from pydantic import Field
from .baseinvocation import BaseInvocation, InvocationContext, InvocationConfig
from ..models.image import ImageField, ResourceOrigin, ImageCategory
from .image import ImageOutput
class ResizeInvocation(BaseInvocation):
'''Resizes an image'''
type: Literal['resize'] = 'resize'
# Inputs
image: Union[ImageField, None] = Field(description="The input image", default=None)
width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
class Config(InvocationConfig):
schema_extra: {
ui: {
tags: ['resize', 'image'],
title: ['My Custom Resize']
}
}
def invoke(self, context: InvocationContext) -> ImageOutput:
# Load the image using InvokeAI's predefined Image Service.
image = context.services.images.get_pil_image(self.image.image_origin, self.image.image_name)
# Resizing the image
# Because we used the above service, we already have a PIL image. So we can simply resize.
resized_image = image.resize((self.width, self.height))
# Preparing the image for output using InvokeAI's predefined Image Service.
output_image = context.services.images.create(
image=resized_image,
image_origin=ResourceOrigin.INTERNAL,
image_category=ImageCategory.GENERAL,
node_id=self.id,
session_id=context.graph_execution_state_id,
is_intermediate=self.is_intermediate,
)
# Returning the Image
return ImageOutput(
image=ImageField(
image_name=output_image.image_name,
image_origin=output_image.image_origin,
),
width=output_image.width,
height=output_image.height,
)
```
We now customized our code to let the frontend know that our Invocation falls
under `resize` and `image` categories. So when the user searches for these
particular words, our Invocation will show up too.
We also set a custom title for our Invocation. So instead of being called
`Resize`, it will be called `My Custom Resize`.
As simple as that.
As time goes by, InvokeAI will further improve and add more customizability for
Invocation configuration. We will have more documentation regarding this at a
later time.
# **[TODO]**
## Custom Components For Frontend
Every backend input type should have a corresponding frontend component so the
UI knows what to render when you use a particular field type.
If you are using existing field types, we already have components for those. So
you don't have to worry about creating anything new. But this might not always
be the case. Sometimes you might want to create new field types and have the
frontend UI deal with it in a different way.
This is where we venture into the world of React and Javascript and create our
own new components for our Invocations. Do not fear the world of JS. It's
actually pretty straightforward.
Let us create a new component for our custom color field we created above. When
we use a color field, let us say we want the UI to display a color picker for
the user to pick from rather than entering values. That is what we will build
now.
---
# OLD -- TO BE DELETED OR MOVED LATER
---
## Creating a new invocation

10
docs/features/CLI.md → docs/deprecated/CLI.md
View File

@ -205,14 +205,14 @@ Here are the invoke> command that apply to txt2img:
| `--seamless` | | `False` | Activate seamless tiling for interesting effects |
| `--seamless_axes` | | `x,y` | Specify which axes to use circular convolution on. |
| `--log_tokenization` | `-t` | `False` | Display a color-coded list of the parsed tokens derived from the prompt |
| `--skip_normalization` | `-x` | `False` | Weighted subprompts will not be normalized. See [Weighted Prompts](./OTHER.md#weighted-prompts) |
| `--skip_normalization` | `-x` | `False` | Weighted subprompts will not be normalized. See [Weighted Prompts](../features/OTHER.md#weighted-prompts) |
| `--upscale <int> <float>` | `-U <int> <float>` | `-U 1 0.75` | Upscale image by magnification factor (2, 4), and set strength of upscaling (0.0-1.0). If strength not set, will default to 0.75. |
| `--facetool_strength <float>` | `-G <float> ` | `-G0` | Fix faces (defaults to using the GFPGAN algorithm); argument indicates how hard the algorithm should try (0.0-1.0) |
| `--facetool <name>` | `-ft <name>` | `-ft gfpgan` | Select face restoration algorithm to use: gfpgan, codeformer |
| `--codeformer_fidelity` | `-cf <float>` | `0.75` | Used along with CodeFormer. Takes values between 0 and 1. 0 produces high quality but low accuracy. 1 produces high accuracy but low quality |
| `--save_original` | `-save_orig` | `False` | When upscaling or fixing faces, this will cause the original image to be saved rather than replaced. |
| `--variation <float>` | `-v<float>` | `0.0` | Add a bit of noise (0.0=none, 1.0=high) to the image in order to generate a series of variations. Usually used in combination with `-S<seed>` and `-n<int>` to generate a series a riffs on a starting image. See [Variations](./VARIATIONS.md). |
| `--with_variations <pattern>` | | `None` | Combine two or more variations. See [Variations](./VARIATIONS.md) for now to use this. |
| `--variation <float>` | `-v<float>` | `0.0` | Add a bit of noise (0.0=none, 1.0=high) to the image in order to generate a series of variations. Usually used in combination with `-S<seed>` and `-n<int>` to generate a series a riffs on a starting image. See [Variations](../features/VARIATIONS.md). |
| `--with_variations <pattern>` | | `None` | Combine two or more variations. See [Variations](../features/VARIATIONS.md) for now to use this. |
| `--save_intermediates <n>` | | `None` | Save the image from every nth step into an "intermediates" folder inside the output directory |
| `--h_symmetry_time_pct <float>` | | `None` | Create symmetry along the X axis at the desired percent complete of the generation process. (Must be between 0.0 and 1.0; set to a very small number like 0.0001 for just after the first step of generation.) |
| `--v_symmetry_time_pct <float>` | | `None` | Create symmetry along the Y axis at the desired percent complete of the generation process. (Must be between 0.0 and 1.0; set to a very small number like 0.0001 for just after the first step of generation.) |
@ -257,7 +257,7 @@ additional options:
by `-M`. You may also supply just a single initial image with the areas
to overpaint made transparent, but you must be careful not to destroy
the pixels underneath when you create the transparent areas. See
[Inpainting](./INPAINTING.md) for details.
[Inpainting](INPAINTING.md) for details.
inpainting accepts all the arguments used for txt2img and img2img, as well as
the --mask (-M) and --text_mask (-tm) arguments:
@ -297,7 +297,7 @@ invoke> a piece of cake -I /path/to/breakfast.png -tm bagel 0.6
You can load and use hundreds of community-contributed Textual
Inversion models just by typing the appropriate trigger phrase. Please
see [Concepts Library](CONCEPTS.md) for more details.
see [Concepts Library](../features/CONCEPTS.md) for more details.
## Other Commands

0
docs/features/EMBIGGEN.md → docs/deprecated/EMBIGGEN.md
View File

0
docs/features/INPAINTING.md → docs/deprecated/INPAINTING.md
View File

0
docs/features/OUTPAINTING.md → docs/deprecated/OUTPAINTING.md
View File

117
docs/features/CONCEPTS.md
View File

@ -1,9 +1,12 @@
---
title: Concepts Library
title: Concepts
---
# :material-library-shelves: The Hugging Face Concepts Library and Importing Textual Inversion files
With the advances in research, many new capabilities are available to customize the knowledge and understanding of novel concepts not originally contained in the base model.
## Using Textual Inversion Files
Textual inversion (TI) files are small models that customize the output of
@ -12,18 +15,16 @@ and artistic styles. They are also known as "embeds" in the machine learning
world.
Each TI file introduces one or more vocabulary terms to the SD model. These are
known in InvokeAI as "triggers." Triggers are often, but not always, denoted
using angle brackets as in "&lt;trigger-phrase&gt;". The two most common type of
known in InvokeAI as "triggers." Triggers are denoted using angle brackets
as in "&lt;trigger-phrase&gt;". The two most common type of
TI files that you'll encounter are `.pt` and `.bin` files, which are produced by
different TI training packages. InvokeAI supports both formats, but its
[built-in TI training system](TEXTUAL_INVERSION.md) produces `.pt`.
[built-in TI training system](TRAINING.md) produces `.pt`.
The [Hugging Face company](https://huggingface.co/sd-concepts-library) has
amassed a large ligrary of &gt;800 community-contributed TI files covering a
broad range of subjects and styles. InvokeAI has built-in support for this
library which downloads and merges TI files automatically upon request. You can
also install your own or others' TI files by placing them in a designated
directory.
broad range of subjects and styles. You can also install your own or others' TI files
by placing them in the designated directory for the compatible model type
### An Example
@ -41,91 +42,43 @@ You can also combine styles and concepts:
| :--------------------------------------------------------: |
| ![](../assets/concepts/image5.png) |
</figure>
## Using a Hugging Face Concept
!!! warning "Authenticating to HuggingFace"
Some concepts require valid authentication to HuggingFace. Without it, they will not be downloaded
and will be silently ignored.
If you used an installer to install InvokeAI, you may have already set a HuggingFace token.
If you skipped this step, you can:
- run the InvokeAI configuration script again (if you used a manual installer): `invokeai-configure`
- set one of the `HUGGINGFACE_TOKEN` or `HUGGING_FACE_HUB_TOKEN` environment variables to contain your token
Finally, if you already used any HuggingFace library on your computer, you might already have a token
in your local cache. Check for a hidden `.huggingface` directory in your home folder. If it
contains a `token` file, then you are all set.
Hugging Face TI concepts are downloaded and installed automatically as you
require them. This requires your machine to be connected to the Internet. To
find out what each concept is for, you can browse the
[Hugging Face concepts library](https://huggingface.co/sd-concepts-library) and
look at examples of what each concept produces.
When you have an idea of a concept you wish to try, go to the command-line
client (CLI) and type a `<` character and the beginning of the Hugging Face
concept name you wish to load. Press ++tab++, and the CLI will show you all
matching concepts. You can also type `<` and hit ++tab++ to get a listing of all
~800 concepts, but be prepared to scroll up to see them all! If there is more
than one match you can continue to type and ++tab++ until the concept is
completed.
!!! example
if you type in `<x` and hit ++tab++, you'll be prompted with the completions:
```py
<xatu2> <xatu> <xbh> <xi> <xidiversity> <xioboma> <xuna> <xyz>
```
Now type `id` and press ++tab++. It will be autocompleted to `<xidiversity>`
because this is a unique match.
Finish your prompt and generate as usual. You may include multiple concept terms
in the prompt.
If you have never used this concept before, you will see a message that the TI
model is being downloaded and installed. After this, the concept will be saved
locally (in the `models/sd-concepts-library` directory) for future use.
Several steps happen during downloading and installation, including a scan of
the file for malicious code. Should any errors occur, you will be warned and the
concept will fail to load. Generation will then continue treating the trigger
term as a normal string of characters (e.g. as literal `<ghibli-face>`).
You can also use `<concept-names>` in the WebGUI's prompt textbox. There is no
autocompletion at this time.
## Installing your Own TI Files
You may install any number of `.pt` and `.bin` files simply by copying them into
the `embeddings` directory of the InvokeAI runtime directory (usually `invokeai`
in your home directory). You may create subdirectories in order to organize the
files in any way you wish. Be careful not to overwrite one file with another.
the `embedding` directory of the corresponding InvokeAI models directory (usually `invokeai`
in your home directory). For example, you can simply move a Stable Diffusion 1.5 embedding file to
the `sd-1/embedding` folder. Be careful not to overwrite one file with another.
For example, TI files generated by the Hugging Face toolkit share the named
`learned_embedding.bin`. You can use subdirectories to keep them distinct.
`learned_embedding.bin`. You can rename these, or use subdirectories to keep them distinct.
At startup time, InvokeAI will scan the `embeddings` directory and load any TI
files it finds there. At startup you will see a message similar to this one:
At startup time, InvokeAI will scan the various `embedding` directories and load any TI
files it finds there for compatible models. At startup you will see a message similar to this one:
```bash
>> Current embedding manager terms: *, <HOI4-Leader>, <princess-knight>
>> Current embedding manager terms: <HOI4-Leader>, <princess-knight>
```
To use these when generating, simply type the `<` key in your prompt to open the Textual Inversion WebUI and
select the embedding you'd like to use. This UI has type-ahead support, so you can easily find supported embeddings.
Note the `*` trigger term. This is a placeholder term that many early TI
tutorials taught people to use rather than a more descriptive term.
Unfortunately, if you have multiple TI files that all use this term, only the
first one loaded will be triggered by use of the term.
## Using LoRAs
To avoid this problem, you can use the `merge_embeddings.py` script to merge two
or more TI files together. If it encounters a collision of terms, the script
will prompt you to select new terms that do not collide. See
[Textual Inversion](TEXTUAL_INVERSION.md) for details.
LoRA files are models that customize the output of Stable Diffusion image generation.
Larger than embeddings, but much smaller than full models, they augment SD with improved
understanding of subjects and artistic styles.
## Further Reading
Unlike TI files, LoRAs do not introduce novel vocabulary into the model's known tokens. Instead,
LoRAs augment the model's weights that are applied to generate imagery. LoRAs may be supplied
with a "trigger" word that they have been explicitly trained on, or may simply apply their
effect without being triggered.
LoRAs are typically stored in .safetensors files, which are the most secure way to store and transmit
these types of weights. You may install any number of `.safetensors` LoRA files simply by copying them into
the `lora` directory of the corresponding InvokeAI models directory (usually `invokeai`
in your home directory). For example, you can simply move a Stable Diffusion 1.5 LoRA file to
the `sd-1/lora` folder.
To use these when generating, open the LoRA menu item in the options panel, select the LoRAs you want to apply
and ensure that they have the appropriate weight recommended by the model provider. Typically, most LoRAs perform best at a weight of .75-1.
Please see [the repository](https://github.com/rinongal/textual_inversion) and
associated paper for details and limitations.

287
docs/features/CONFIGURATION.md Normal file
View File

@ -0,0 +1,287 @@
---
title: Configuration
---
# :material-tune-variant: InvokeAI Configuration
## Intro
InvokeAI has numerous runtime settings which can be used to adjust
many aspects of its operations, including the location of files and
directories, memory usage, and performance. These settings can be
viewed and customized in several ways:
1. By editing settings in the `invokeai.yaml` file.
2. By setting environment variables.
3. On the command-line, when InvokeAI is launched.
In addition, the most commonly changed settings are accessible
graphically via the `invokeai-configure` script.
### How the Configuration System Works
When InvokeAI is launched, the very first thing it needs to do is to
find its "root" directory, which contains its configuration files,
installed models, its database of images, and the folder(s) of
generated images themselves. In this document, the root directory will
be referred to as ROOT.
#### Finding the Root Directory
To find its root directory, InvokeAI uses the following recipe:
1. It first looks for the argument `--root <path>` on the command line
it was launched from, and uses the indicated path if present.
2. Next it looks for the environment variable INVOKEAI_ROOT, and uses
the directory path found there if present.
3. If neither of these are present, then InvokeAI looks for the
folder containing the `.venv` Python virtual environment directory for
the currently active environment. This directory is checked for files
expected inside the InvokeAI root before it is used.
4. Finally, InvokeAI looks for a directory in the current user's home
directory named `invokeai`.
#### Reading the InvokeAI Configuration File
Once the root directory has been located, InvokeAI looks for a file
named `ROOT/invokeai.yaml`, and if present reads configuration values
from it. The top of this file looks like this:
```
InvokeAI:
Web Server:
host: localhost
port: 9090
allow_origins: []
allow_credentials: true
allow_methods:
- '*'
allow_headers:
- '*'
Features:
esrgan: true
internet_available: true
log_tokenization: false
nsfw_checker: false
patchmatch: true
restore: true
...
```
This lines in this file are used to establish default values for
Invoke's settings. In the above fragment, the Web Server's listening
port is set to 9090 by the `port` setting.
You can edit this file with a text editor such as "Notepad" (do not
use Word or any other word processor). When editing, be careful to
maintain the indentation, and do not add extraneous text, as syntax
errors will prevent InvokeAI from launching. A basic guide to the
format of YAML files can be found
[here](https://circleci.com/blog/what-is-yaml-a-beginner-s-guide/).
You can fix a broken `invokeai.yaml` by deleting it and running the
configuration script again -- option [7] in the launcher, "Re-run the
configure script".
#### Reading Environment Variables
Next InvokeAI looks for defined environment variables in the format
`INVOKEAI_<setting_name>`, for example `INVOKEAI_port`. Environment
variable values take precedence over configuration file variables. On
a Macintosh system, for example, you could change the port that the
web server listens on by setting the environment variable this way:
```
export INVOKEAI_port=8000
invokeai-web
```
Please check out these
[Macintosh](https://phoenixnap.com/kb/set-environment-variable-mac)
and
[Windows](https://phoenixnap.com/kb/windows-set-environment-variable)
guides for setting temporary and permanent environment variables.
#### Reading the Command Line
Lastly, InvokeAI takes settings from the command line, which override
everything else. The command-line settings have the same name as the
corresponding configuration file settings, preceded by a `--`, for
example `--port 8000`.
If you are using the launcher (`invoke.sh` or `invoke.bat`) to launch
InvokeAI, then just pass the command-line arguments to the launcher:
```
invoke.bat --port 8000 --host 0.0.0.0
```
The arguments will be applied when you select the web server option
(and the other options as well).
If, on the other hand, you prefer to launch InvokeAI directly from the
command line, you would first activate the virtual environment (known
as the "developer's console" in the launcher), and run `invokeai-web`:
```
> C:\Users\Fred\invokeai\.venv\scripts\activate
(.venv) > invokeai-web --port 8000 --host 0.0.0.0
```
You can get a listing and brief instructions for each of the
command-line options by giving the `--help` argument:
```
(.venv) > invokeai-web --help
usage: InvokeAI [-h] [--host HOST] [--port PORT] [--allow_origins [ALLOW_ORIGINS ...]] [--allow_credentials | --no-allow_credentials]
[--allow_methods [ALLOW_METHODS ...]] [--allow_headers [ALLOW_HEADERS ...]] [--esrgan | --no-esrgan]
[--internet_available | --no-internet_available] [--log_tokenization | --no-log_tokenization]
[--nsfw_checker | --no-nsfw_checker] [--patchmatch | --no-patchmatch] [--restore | --no-restore]
[--always_use_cpu | --no-always_use_cpu] [--free_gpu_mem | --no-free_gpu_mem] [--max_cache_size MAX_CACHE_SIZE]
[--max_vram_cache_size MAX_VRAM_CACHE_SIZE] [--precision {auto,float16,float32,autocast}]
[--sequential_guidance | --no-sequential_guidance] [--xformers_enabled | --no-xformers_enabled]
[--tiled_decode | --no-tiled_decode] [--root ROOT] [--autoimport_dir AUTOIMPORT_DIR] [--lora_dir LORA_DIR]
[--embedding_dir EMBEDDING_DIR] [--controlnet_dir CONTROLNET_DIR] [--conf_path CONF_PATH] [--models_dir MODELS_DIR]
[--legacy_conf_dir LEGACY_CONF_DIR] [--db_dir DB_DIR] [--outdir OUTDIR] [--from_file FROM_FILE]
[--use_memory_db | --no-use_memory_db] [--model MODEL] [--log_handlers [LOG_HANDLERS ...]]
[--log_format {plain,color,syslog,legacy}] [--log_level {debug,info,warning,error,critical}]
...
```
## The Configuration Settings
The configuration settings are divided into several distinct
groups in `invokeia.yaml`:
### Web Server
| Setting | Default Value | Description |
|----------|----------------|--------------|
| `host` | `localhost` | Name or IP address of the network interface that the web server will listen on |
| `port` | `9090` | Network port number that the web server will listen on |
| `allow_origins` | `[]` | A list of host names or IP addresses that are allowed to connect to the InvokeAI API in the format `['host1','host2',...]` |
| `allow_credentials | `true` | Require credentials for a foreign host to access the InvokeAI API (don't change this) |
| `allow_methods` | `*` | List of HTTP methods ("GET", "POST") that the web server is allowed to use when accessing the API |
| `allow_headers` | `*` | List of HTTP headers that the web server will accept when accessing the API |
The documentation for InvokeAI's API can be accessed by browsing to the following URL: [http://localhost:9090/docs].
### Features
These configuration settings allow you to enable and disable various InvokeAI features:
| Setting | Default Value | Description |
|----------|----------------|--------------|
| `esrgan` | `true` | Activate the ESRGAN upscaling options|
| `internet_available` | `true` | When a resource is not available locally, try to fetch it via the internet |
| `log_tokenization` | `false` | Before each text2image generation, print a color-coded representation of the prompt to the console; this can help understand why a prompt is not working as expected |
| `nsfw_checker` | `true` | Activate the NSFW checker to blur out risque images |
| `patchmatch` | `true` | Activate the "patchmatch" algorithm for improved inpainting |
| `restore` | `true` | Activate the facial restoration features (DEPRECATED; restoration features will be removed in 3.0.0) |
### Memory/Performance
These options tune InvokeAI's memory and performance characteristics.
| Setting | Default Value | Description |
|----------|----------------|--------------|
| `always_use_cpu` | `false` | Use the CPU to generate images, even if a GPU is available |
| `free_gpu_mem` | `false` | Aggressively free up GPU memory after each operation; this will allow you to run in low-VRAM environments with some performance penalties |
| `max_cache_size` | `6` | Amount of CPU RAM (in GB) to reserve for caching models in memory; more cache allows you to keep models in memory and switch among them quickly |
| `max_vram_cache_size` | `2.75` | Amount of GPU VRAM (in GB) to reserve for caching models in VRAM; more cache speeds up generation but reduces the size of the images that can be generated. This can be set to zero to maximize the amount of memory available for generation. |
| `precision` | `auto` | Floating point precision. One of `auto`, `float16` or `float32`. `float16` will consume half the memory of `float32` but produce slightly lower-quality images. The `auto` setting will guess the proper precision based on your video card and operating system |
| `sequential_guidance` | `false` | Calculate guidance in serial rather than in parallel, lowering memory requirements at the cost of some performance loss |
| `xformers_enabled` | `true` | If the x-formers memory-efficient attention module is installed, activate it for better memory usage and generation speed|
| `tiled_decode` | `false` | If true, then during the VAE decoding phase the image will be decoded a section at a time, reducing memory consumption at the cost of a performance hit |
### Paths
These options set the paths of various directories and files used by
InvokeAI. Relative paths are interpreted relative to INVOKEAI_ROOT, so
if INVOKEAI_ROOT is `/home/fred/invokeai` and the path is
`autoimport/main`, then the corresponding directory will be located at
`/home/fred/invokeai/autoimport/main`.
| Setting | Default Value | Description |
|----------|----------------|--------------|
| `autoimport_dir` | `autoimport/main` | At startup time, read and import any main model files found in this directory |
| `lora_dir` | `autoimport/lora` | At startup time, read and import any LoRA/LyCORIS models found in this directory |
| `embedding_dir` | `autoimport/embedding` | At startup time, read and import any textual inversion (embedding) models found in this directory |
| `controlnet_dir` | `autoimport/controlnet` | At startup time, read and import any ControlNet models found in this directory |
| `conf_path` | `configs/models.yaml` | Location of the `models.yaml` model configuration file |
| `models_dir` | `models` | Location of the directory containing models installed by InvokeAI's model manager |
| `legacy_conf_dir` | `configs/stable-diffusion` | Location of the directory containing the .yaml configuration files for legacy checkpoint models |
| `db_dir` | `databases` | Location of the directory containing InvokeAI's image, schema and session database |
| `outdir` | `outputs` | Location of the directory in which the gallery of generated and uploaded images will be stored |
| `use_memory_db` | `false` | Keep database information in memory rather than on disk; this will not preserve image gallery information across restarts |
Note that the autoimport directories will be searched recursively,
allowing you to organize the models into folders and subfolders in any
way you wish. In addition, while we have split up autoimport
directories by the type of model they contain, this isn't
necessary. You can combine different model types in the same folder
and InvokeAI will figure out what they are. So you can easily use just
one autoimport directory by commenting out the unneeded paths:
```
Paths:
autoimport_dir: autoimport
# lora_dir: null
# embedding_dir: null
# controlnet_dir: null
```
### Logging
These settings control the information, warning, and debugging
messages printed to the console log while InvokeAI is running:
| Setting | Default Value | Description |
|----------|----------------|--------------|
| `log_handlers` | `console` | This controls where log messages are sent, and can be a list of one or more destinations. Values include `console`, `file`, `syslog` and `http`. These are described in more detail below |
| `log_format` | `color` | This controls the formatting of the log messages. Values are `plain`, `color`, `legacy` and `syslog` |
| `log_level` | `debug` | This filters messages according to the level of severity and can be one of `debug`, `info`, `warning`, `error` and `critical`. For example, setting to `warning` will display all messages at the warning level or higher, but won't display "debug" or "info" messages |
Several different log handler destinations are available, and multiple destinations are supported by providing a list:
```
log_handlers:
- console
- syslog=localhost
- file=/var/log/invokeai.log
```
* `console` is the default. It prints log messages to the command-line window from which InvokeAI was launched.
* `syslog` is only available on Linux and Macintosh systems. It uses
the operating system's "syslog" facility to write log file entries
locally or to a remote logging machine. `syslog` offers a variety
of configuration options:
```
syslog=/dev/log` - log to the /dev/log device
syslog=localhost` - log to the network logger running on the local machine
syslog=localhost:512` - same as above, but using a non-standard port
syslog=fredserver,facility=LOG_USER,socktype=SOCK_DRAM`
- Log to LAN-connected server "fredserver" using the facility LOG_USER and datagram packets.
```
* `http` can be used to log to a remote web server. The server must be
properly configured to receive and act on log messages. The option
accepts the URL to the web server, and a `method` argument
indicating whether the message should be submitted using the GET or
POST method.
```
http=http://my.server/path/to/logger,method=POST
```
The `log_format` option provides several alternative formats:
* `color` - default format providing time, date and a message, using text colors to distinguish different log severities
* `plain` - same as above, but monochrome text only
* `syslog` - the log level and error message only, allowing the syslog system to attach the time and date
* `legacy` - a format similar to the one used by the legacy 2.3 InvokeAI releases.

92
docs/features/CONTROLNET.md Normal file
View File

@ -0,0 +1,92 @@
---
title: ControlNet
---
# :material-loupe: ControlNet
## ControlNet
ControlNet
ControlNet is a powerful set of features developed by the open-source community (notably, Stanford researcher [**@ilyasviel**](https://github.com/lllyasviel)) that allows you to apply a secondary neural network model to your image generation process in Invoke.
With ControlNet, you can get more control over the output of your image generation, providing you with a way to direct the network towards generating images that better fit your desired style or outcome.
### How it works
ControlNet works by analyzing an input image, pre-processing that image to identify relevant information that can be interpreted by each specific ControlNet model, and then inserting that control information into the generation process. This can be used to adjust the style, composition, or other aspects of the image to better achieve a specific result.
### Models
As part of the model installation, ControlNet models can be selected including a variety of pre-trained models that have been added to achieve different effects or styles in your generated images. Further ControlNet models may require additional code functionality to also be incorporated into Invoke's Invocations folder. You should expect to follow any installation instructions for ControlNet models loaded outside the default models provided by Invoke. The default models include:
**Canny**:
When the Canny model is used in ControlNet, Invoke will attempt to generate images that match the edges detected.
Canny edge detection works by detecting the edges in an image by looking for abrupt changes in intensity. It is known for its ability to detect edges accurately while reducing noise and false edges, and the preprocessor can identify more information by decreasing the thresholds.
**M-LSD**:
M-LSD is another edge detection algorithm used in ControlNet. It stands for Multi-Scale Line Segment Detector.
It detects straight line segments in an image by analyzing the local structure of the image at multiple scales. It can be useful for architectural imagery, or anything where straight-line structural information is needed for the resulting output.
**Lineart**:
The Lineart model in ControlNet generates line drawings from an input image. The resulting pre-processed image is a simplified version of the original, with only the outlines of objects visible.The Lineart model in ControlNet is known for its ability to accurately capture the contours of the objects in an input sketch.
**Lineart Anime**:
A variant of the Lineart model that generates line drawings with a distinct style inspired by anime and manga art styles.
**Depth**:
A model that generates depth maps of images, allowing you to create more realistic 3D models or to simulate depth effects in post-processing.
**Normal Map (BAE):**
A model that generates normal maps from input images, allowing for more realistic lighting effects in 3D rendering.
**Image Segmentation**:
A model that divides input images into segments or regions, each of which corresponds to a different object or part of the image. (More details coming soon)
**Openpose**:
The OpenPose control model allows for the identification of the general pose of a character by pre-processing an existing image with a clear human structure. With advanced options, Openpose can also detect the face or hands in the image.
**Mediapipe Face**:
The MediaPipe Face identification processor is able to clearly identify facial features in order to capture vivid expressions of human faces.
**Tile (experimental)**:
The Tile model fills out details in the image to match the image, rather than the prompt. The Tile Model is a versatile tool that offers a range of functionalities. Its primary capabilities can be boiled down to two main behaviors:
- It can reinterpret specific details within an image and create fresh, new elements.
- It has the ability to disregard global instructions if there's a discrepancy between them and the local context or specific parts of the image. In such cases, it uses the local context to guide the process.
The Tile Model can be a powerful tool in your arsenal for enhancing image quality and details. If there are undesirable elements in your images, such as blurriness caused by resizing, this model can effectively eliminate these issues, resulting in cleaner, crisper images. Moreover, it can generate and add refined details to your images, improving their overall quality and appeal.
**Pix2Pix (experimental)**
With Pix2Pix, you can input an image into the controlnet, and then "instruct" the model to change it using your prompt. For example, you can say "Make it winter" to add more wintry elements to a scene.
**Inpaint**: Coming Soon - Currently this model is available but not functional on the Canvas. An upcoming release will provide additional capabilities for using this model when inpainting.
Each of these models can be adjusted and combined with other ControlNet models to achieve different results, giving you even more control over your image generation process.
## Using ControlNet
To use ControlNet, you can simply select the desired model and adjust both the ControlNet and Pre-processor settings to achieve the desired result. You can also use multiple ControlNet models at the same time, allowing you to achieve even more complex effects or styles in your generated images.
Each ControlNet has two settings that are applied to the ControlNet.
Weight - Strength of the Controlnet model applied to the generation for the section, defined by start/end.
Start/End - 0 represents the start of the generation, 1 represents the end. The Start/end setting controls what steps during the generation process have the ControlNet applied.
Additionally, each ControlNet section can be expanded in order to manipulate settings for the image pre-processor that adjusts your uploaded image before using it in when you Invoke.

110
docs/features/IMG2IMG.md
View File

@ -4,86 +4,13 @@ title: Image-to-Image
# :material-image-multiple: Image-to-Image
Both the Web and command-line interfaces provide an "img2img" feature
that lets you seed your creations with an initial drawing or
photo. This is a really cool feature that tells stable diffusion to
build the prompt on top of the image you provide, preserving the
original's basic shape and layout.
InvokeAI provides an "img2img" feature that lets you seed your
creations with an initial drawing or photo. This is a really cool
feature that tells stable diffusion to build the prompt on top of the
image you provide, preserving the original's basic shape and layout.
See the [WebUI Guide](WEB.md) for a walkthrough of the img2img feature
in the InvokeAI web server. This document describes how to use img2img
in the command-line tool.
## Basic Usage
Launch the command-line client by launching `invoke.sh`/`invoke.bat`
and choosing option (1). Alternative, activate the InvokeAI
environment and issue the command `invokeai`.
Once the `invoke> ` prompt appears, you can start an img2img render by
pointing to a seed file with the `-I` option as shown here:
!!! example ""
```commandline
tree on a hill with a river, nature photograph, national geographic -I./test-pictures/tree-and-river-sketch.png -f 0.85
```
<figure markdown>
| original image | generated image |
| :------------: | :-------------: |
| ![original-image](https://user-images.githubusercontent.com/50542132/193946000-c42a96d8-5a74-4f8a-b4c3-5213e6cadcce.png){ width=320 } | ![generated-image](https://user-images.githubusercontent.com/111189/194135515-53d4c060-e994-4016-8121-7c685e281ac9.png){ width=320 } |
</figure>
The `--init_img` (`-I`) option gives the path to the seed picture. `--strength`
(`-f`) controls how much the original will be modified, ranging from `0.0` (keep
the original intact), to `1.0` (ignore the original completely). The default is
`0.75`, and ranges from `0.25-0.90` give interesting results. Other relevant
options include `-C` (classification free guidance scale), and `-s` (steps).
Unlike `txt2img`, adding steps will continuously change the resulting image and
it will not converge.
You may also pass a `-v<variation_amount>` option to generate `-n<iterations>`
count variants on the original image. This is done by passing the first
generated image back into img2img the requested number of times. It generates
interesting variants.
Note that the prompt makes a big difference. For example, this slight variation
on the prompt produces a very different image:
<figure markdown>
![](https://user-images.githubusercontent.com/111189/194135220-16b62181-b60c-4248-8989-4834a8fd7fbd.png){ width=320 }
<caption markdown>photograph of a tree on a hill with a river</caption>
</figure>
!!! tip
When designing prompts, think about how the images scraped from the internet were
captioned. Very few photographs will be labeled "photograph" or "photorealistic."
They will, however, be captioned with the publication, photographer, camera model,
or film settings.
If the initial image contains transparent regions, then Stable Diffusion will
only draw within the transparent regions, a process called
[`inpainting`](./INPAINTING.md#creating-transparent-regions-for-inpainting).
However, for this to work correctly, the color information underneath the
transparent needs to be preserved, not erased.
!!! warning "**IMPORTANT ISSUE** "
`img2img` does not work properly on initial images smaller
than 512x512. Please scale your image to at least 512x512 before using it.
Larger images are not a problem, but may run out of VRAM on your GPU card. To
fix this, use the --fit option, which downscales the initial image to fit within
the box specified by width x height:
```
tree on a hill with a river, national geographic -I./test-pictures/big-sketch.png -H512 -W512 --fit
```
## How does it actually work, though?
For a walkthrough of using Image-to-Image in the Web UI, see [InvokeAI
Web Server](./WEB.md#image-to-image).
The main difference between `img2img` and `prompt2img` is the starting point.
While `prompt2img` always starts with pure gaussian noise and progressively
@ -99,10 +26,6 @@ seed `1592514025` develops something like this:
!!! example ""
```bash
invoke> "fire" -s10 -W384 -H384 -S1592514025
```
<figure markdown>
![latent steps](../assets/img2img/000019.steps.png){ width=720 }
</figure>
@ -157,17 +80,8 @@ Diffusion has less chance to refine itself, so the result ends up inheriting all
the problems of my bad drawing.
If you want to try this out yourself, all of these are using a seed of
`1592514025` with a width/height of `384`, step count `10`, the default sampler
(`k_lms`), and the single-word prompt `"fire"`:
```bash
invoke> "fire" -s10 -W384 -H384 -S1592514025 -I /tmp/fire-drawing.png --strength 0.7
```
The code for rendering intermediates is on my (damian0815's) branch
[document-img2img](https://github.com/damian0815/InvokeAI/tree/document-img2img) -
run `invoke.py` and check your `outputs/img-samples/intermediates` folder while
generating an image.
`1592514025` with a width/height of `384`, step count `10`, the
`k_lms` sampler, and the single-word prompt `"fire"`.
### Compensating for the reduced step count
@ -180,10 +94,6 @@ give each generation 20 steps.
Here's strength `0.4` (note step count `50`, which is `20 ÷ 0.4` to make sure SD
does `20` steps from my image):
```bash
invoke> "fire" -s50 -W384 -H384 -S1592514025 -I /tmp/fire-drawing.png -f 0.4
```
<figure markdown>
![000035.1592514025](../assets/img2img/000035.1592514025.png)
</figure>
@ -191,10 +101,6 @@ invoke> "fire" -s50 -W384 -H384 -S1592514025 -I /tmp/fire-drawing.png -f 0.4
and here is strength `0.7` (note step count `30`, which is roughly `20 ÷ 0.7` to
make sure SD does `20` steps from my image):
```commandline
invoke> "fire" -s30 -W384 -H384 -S1592514025 -I /tmp/fire-drawing.png -f 0.7
```
<figure markdown>
![000046.1592514025](../assets/img2img/000046.1592514025.png)
</figure>

3
docs/features/MODEL_MERGING.md
View File

@ -71,6 +71,3 @@ under the selected name and register it with InvokeAI.
use InvokeAI conventions - only alphanumeric letters and the
characters ".+-".
## Caveats
This is a new script and may contain bugs.

206
docs/features/NODES.md Normal file
View File

@ -0,0 +1,206 @@
# Nodes Editor (Experimental)
🚨
*The node editor is experimental. We've made it accessible because we use it to develop the application, but we have not addressed the many known rough edges. It's very easy to shoot yourself in the foot, and we cannot offer support for it until it sees full release (ETA v3.1). Everything is subject to change without warning.*
🚨
The nodes editor is a blank canvas allowing for the use of individual functions and image transformations to control the image generation workflow. The node processing flow is usually done from left (inputs) to right (outputs), though linearity can become abstracted the more complex the node graph becomes. Nodes inputs and outputs are connected by dragging connectors from node to node.
To better understand how nodes are used, think of how an electric power bar works. It takes in one input (electricity from a wall outlet) and passes it to multiple devices through multiple outputs. Similarly, a node could have multiple inputs and outputs functioning at the same (or different) time, but all node outputs pass information onward like a power bar passes electricity. Not all outputs are compatible with all inputs, however - Each node has different constraints on how it is expecting to input/output information. In general, node outputs are colour-coded to match compatible inputs of other nodes.
## Anatomy of a Node
Individual nodes are made up of the following:
- Inputs: Edge points on the left side of the node window where you connect outputs from other nodes.
- Outputs: Edge points on the right side of the node window where you connect to inputs on other nodes.
- Options: Various options which are either manually configured, or overridden by connecting an output from another node to the input.
## Diffusion Overview
Taking the time to understand the diffusion process will help you to understand how to set up your nodes in the nodes editor.
There are two main spaces Stable Diffusion works in: image space and latent space.
Image space represents images in pixel form that you look at. Latent space represents compressed inputs. It’s in latent space that Stable Diffusion processes images. A VAE (Variational Auto Encoder) is responsible for compressing and encoding inputs into latent space, as well as decoding outputs back into image space.
When you generate an image using text-to-image, multiple steps occur in latent space:
1. Random noise is generated at the chosen height and width. The noise’s characteristics are dictated by the chosen (or not chosen) seed. This noise tensor is passed into latent space. We’ll call this noise A.
1. Using a model’s U-Net, a noise predictor examines noise A, and the words tokenized by CLIP from your prompt (conditioning). It generates its own noise tensor to predict what the final image might look like in latent space. We’ll call this noise B.
1. Noise B is subtracted from noise A in an attempt to create a final latent image indicative of the inputs. This step is repeated for the number of sampler steps chosen.
1. The VAE decodes the final latent image from latent space into image space.
image-to-image is a similar process, with only step 1 being different:
1. The input image is decoded from image space into latent space by the VAE. Noise is then added to the input latent image. Denoising Strength dictates how much noise is added, 0 being none, and 1 being all-encompassing. We’ll call this noise A. The process is then the same as steps 2-4 in the text-to-image explanation above.
Furthermore, a model provides the CLIP prompt tokenizer, the VAE, and a U-Net (where noise prediction occurs given a prompt and initial noise tensor).
A noise scheduler (eg. DPM++ 2M Karras) schedules the subtraction of noise from the latent image across the sampler steps chosen (step 3 above). Less noise is usually subtracted at higher sampler steps.
## Node Types (Base Nodes)
| Node <img width=160 align="right"> | Function |
| ---------------------------------- | --------------------------------------------------------------------------------------|
| Add | Adds two numbers |
| CannyImageProcessor | Canny edge detection for ControlNet |
| ClipSkip | Skip layers in clip text_encoder model |
| Collect | Collects values into a collection |
| Prompt (Compel) | Parse prompt using compel package to conditioning |
| ContentShuffleImageProcessor | Applies content shuffle processing to image |
| ControlNet | Collects ControlNet info to pass to other nodes |
| CvInpaint | Simple inpaint using opencv |
| Divide | Divides two numbers |
| DynamicPrompt | Parses a prompt using adieyal/dynamic prompt's random or combinatorial generator |
| FloatLinearRange | Creates a range |
| HedImageProcessor | Applies HED edge detection to image |
| ImageBlur | Blurs an image |
| ImageChannel | Gets a channel from an image |
| ImageCollection | Load a collection of images and provide it as output |
| ImageConvert | Converts an image to a different mode |
| ImageCrop | Crops an image to a specified box. The box can be outside of the image. |
| ImageInverseLerp | Inverse linear interpolation of all pixels of an image |
| ImageLerp | Linear interpolation of all pixels of an image |
| ImageMultiply | Multiplies two images together using `PIL.ImageChops.Multiply()` |
| ImagePaste | Pastes an image into another image |
| ImageProcessor | Base class for invocations that reprocess images for ControlNet |
| ImageResize | Resizes an image to specific dimensions |
| ImageScale | Scales an image by a factor |
| ImageToLatents | Scales latents by a given factor |
| InfillColor | Infills transparent areas of an image with a solid color |
| InfillPatchMatch | Infills transparent areas of an image using the PatchMatch algorithm |
| InfillTile | Infills transparent areas of an image with tiles of the image |
| Inpaint | Generates an image using inpaint |
| Iterate | Iterates over a list of items |
| LatentsToImage | Generates an image from latents |
| LatentsToLatents | Generates latents using latents as base image |
| LeresImageProcessor | Applies leres processing to image |
| LineartAnimeImageProcessor | Applies line art anime processing to image |
| LineartImageProcessor | Applies line art processing to image |
| LoadImage | Load an image and provide it as output |
| Lora Loader | Apply selected lora to unet and text_encoder |
| Model Loader | Loads a main model, outputting its submodels |
| MaskFromAlpha | Extracts the alpha channel of an image as a mask |
| MediapipeFaceProcessor | Applies mediapipe face processing to image |
| MidasDepthImageProcessor | Applies Midas depth processing to image |
| MlsdImageProcessor | Applied MLSD processing to image |
| Multiply | Multiplies two numbers |
| Noise | Generates latent noise |
| NormalbaeImageProcessor | Applies NormalBAE processing to image |
| OpenposeImageProcessor | Applies Openpose processing to image |
| ParamFloat | A float parameter |
| ParamInt | An integer parameter |
| PidiImageProcessor | Applies PIDI processing to an image |
| Progress Image | Displays the progress image in the Node Editor |
| RandomInit | Outputs a single random integer |
| RandomRange | Creates a collection of random numbers |
| Range | Creates a range of numbers from start to stop with step |
| RangeOfSize | Creates a range from start to start + size with step |
| ResizeLatents | Resizes latents to explicit width/height (in pixels). Provided dimensions are floor-divided by 8. |
| RestoreFace | Restores faces in the image |
| ScaleLatents | Scales latents by a given factor |
| SegmentAnythingProcessor | Applies segment anything processing to image |
| ShowImage | Displays a provided image, and passes it forward in the pipeline |
| StepParamEasing | Experimental per-step parameter for easing for denoising steps |
| Subtract | Subtracts two numbers |
| TextToLatents | Generates latents from conditionings |
| TileResampleProcessor | Bass class for invocations that preprocess images for ControlNet |
| Upscale | Upscales an image |
| VAE Loader | Loads a VAE model, outputting a VaeLoaderOutput |
| ZoeDepthImageProcessor | Applies Zoe depth processing to image |
## Node Grouping Concepts
There are several node grouping concepts that can be examined with a narrow focus. These (and other) groupings can be pieced together to make up functional graph setups, and are important to understanding how groups of nodes work together as part of a whole. Note that the screenshots below aren't examples of complete functioning node graphs (see Examples).
### Noise
As described, an initial noise tensor is necessary for the latent diffusion process. As a result, all non-image *ToLatents nodes require a noise node input.
<img width="654" alt="groupsnoise" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/2e8d297e-ad55-4d27-bc93-c119dad2a2c5">
### Conditioning
As described, conditioning is necessary for the latent diffusion process, whether empty or not. As a result, all non-image *ToLatents nodes require positive and negative conditioning inputs. Conditioning is reliant on a CLIP tokenizer provided by the Model Loader node.
<img width="1024" alt="groupsconditioning" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/f8f7ad8a-8d9c-418e-b5ad-1437b774b27e">
### Image Space & VAE
The ImageToLatents node doesn't require a noise node input, but requires a VAE input to convert the image from image space into latent space. In reverse, the LatentsToImage node requires a VAE input to convert from latent space back into image space.
<img width="637" alt="groupsimgvae" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/dd99969c-e0a8-4f78-9b17-3ffe179cef9a">
### Defined & Random Seeds
It is common to want to use both the same seed (for continuity) and random seeds (for variance). To define a seed, simply enter it into the 'Seed' field on a noise node. Conversely, the RandomInt node generates a random integer between 'Low' and 'High', and can be used as input to the 'Seed' edge point on a noise node to randomize your seed.
<img width="922" alt="groupsrandseed" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/af55bc20-60f6-438e-aba5-3ec871443710">
### Control
Control means to guide the diffusion process to adhere to a defined input or structure. Control can be provided as input to non-image *ToLatents nodes from ControlNet nodes. ControlNet nodes usually require an image processor which converts an input image for use with ControlNet.
<img width="805" alt="groupscontrol" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/cc9c5de7-23a7-46c8-bbad-1f3609d999a6">
### LoRA
The Lora Loader node lets you load a LoRA (say that ten times fast) and pass it as output to both the Prompt (Compel) and non-image *ToLatents nodes. A model's CLIP tokenizer is passed through the LoRA into Prompt (Compel), where it affects conditioning. A model's U-Net is also passed through the LoRA into a non-image *ToLatents node, where it affects noise prediction.
<img width="993" alt="groupslora" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/630962b0-d914-4505-b3ea-ccae9b0269da">
### Scaling
Use the ImageScale, ScaleLatents, and Upscale nodes to upscale images and/or latent images. The chosen method differs across contexts. However, be aware that latents are already noisy and compressed at their original resolution; scaling an image could produce more detailed results.
<img width="644" alt="groupsallscale" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/99314f05-dd9f-4b6d-b378-31de55346a13">
### Iteration + Multiple Images as Input
Iteration is a common concept in any processing, and means to repeat a process with given input. In nodes, you're able to use the Iterate node to iterate through collections usually gathered by the Collect node. The Iterate node has many potential uses, from processing a collection of images one after another, to varying seeds across multiple image generations and more. This screenshot demonstrates how to collect several images and pass them out one at a time.
<img width="788" alt="groupsiterate" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/4af5ca27-82c9-4018-8c5b-024d3ee0a121">
### Multiple Image Generation + Random Seeds
Multiple image generation in the node editor is done using the RandomRange node. In this case, the 'Size' field represents the number of images to generate. As RandomRange produces a collection of integers, we need to add the Iterate node to iterate through the collection.
To control seeds across generations takes some care. The first row in the screenshot will generate multiple images with different seeds, but using the same RandomRange parameters across invocations will result in the same group of random seeds being used across the images, producing repeatable results. In the second row, adding the RandomInt node as input to RandomRange's 'Seed' edge point will ensure that seeds are varied across all images across invocations, producing varied results.
<img width="1027" alt="groupsmultigenseeding" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/518d1b2b-fed1-416b-a052-ab06552521b3">
## Examples
With our knowledge of node grouping and the diffusion process, let’s break down some basic graphs in the nodes editor. Note that a node's options can be overridden by inputs from other nodes. These examples aren't strict rules to follow and only demonstrate some basic configurations.
### Basic text-to-image Node Graph
<img width="875" alt="nodest2i" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/17c67720-c376-4db8-94f0-5e00381a61ee">
- Model Loader: A necessity to generating images (as we’ve read above). We choose our model from the dropdown. It outputs a U-Net, CLIP tokenizer, and VAE.
- Prompt (Compel): Another necessity. Two prompt nodes are created. One will output positive conditioning (what you want, ‘dog’), one will output negative (what you don’t want, ‘cat’). They both input the CLIP tokenizer that the Model Loader node outputs.
- Noise: Consider this noise A from step one of the text-to-image explanation above. Choose a seed number, width, and height.
- TextToLatents: This node takes many inputs for converting and processing text & noise from image space into latent space, hence the name TextTo**Latents**. In this setup, it inputs positive and negative conditioning from the prompt nodes for processing (step 2 above). It inputs noise from the noise node for processing (steps 2 & 3 above). Lastly, it inputs a U-Net from the Model Loader node for processing (step 2 above). It outputs latents for use in the next LatentsToImage node. Choose number of sampler steps, CFG scale, and scheduler.
- LatentsToImage: This node takes in processed latents from the TextToLatents node, and the model’s VAE from the Model Loader node which is responsible for decoding latents back into the image space, hence the name LatentsTo**Image**. This node is the last stop, and once the image is decoded, it is saved to the gallery.
### Basic image-to-image Node Graph
<img width="998" alt="nodesi2i" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/3f2c95d5-cee7-4415-9b79-b46ee60a92fe">
- Model Loader: Choose a model from the dropdown.
- Prompt (Compel): Two prompt nodes. One positive (dog), one negative (dog). Same CLIP inputs from the Model Loader node as before.
- ImageToLatents: Upload a source image directly in the node window, via drag'n'drop from the gallery, or passed in as input. The ImageToLatents node inputs the VAE from the Model Loader node to decode the chosen image from image space into latent space, hence the name ImageTo**Latents**. It outputs latents for use in the next LatentsToLatents node. It also outputs the source image's width and height for use in the next Noise node if the final image is to be the same dimensions as the source image.
- Noise: A noise tensor is created with the width and height of the source image, and connected to the next LatentsToLatents node. Notice the width and height fields are overridden by the input from the ImageToLatents width and height outputs.
- LatentsToLatents: The inputs and options are nearly identical to TextToLatents, except that LatentsToLatents also takes latents as an input. Considering our source image is already converted to latents in the last ImageToLatents node, and text + noise are no longer the only inputs to process, we use the LatentsToLatents node.
- LatentsToImage: Like previously, the LatentsToImage node will use the VAE from the Model Loader as input to decode the latents from LatentsToLatents into image space, and save it to the gallery.
### Basic ControlNet Node Graph
<img width="703" alt="nodescontrol" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/b02ded86-ceb4-44a2-9910-e19ad184d471">
- Model Loader
- Prompt (Compel)
- Noise: Width and height of the CannyImageProcessor ControlNet image is passed in to set the dimensions of the noise passed to TextToLatents.
- CannyImageProcessor: The CannyImageProcessor node is used to process the source image being used as a ControlNet. Each ControlNet processor node applies control in different ways, and has some different options to configure. Width and height are passed to noise, as mentioned. The processed ControlNet image is output to the ControlNet node.
- ControlNet: Select the type of control model. In this case, canny is chosen as the CannyImageProcessor was used to generate the ControlNet image. Configure the control node options, and pass the control output to TextToLatents.
- TextToLatents: Similar to the basic text-to-image example, except ControlNet is passed to the control input edge point.
- LatentsToImage

28
docs/features/NSFW.md
View File

@ -31,10 +31,22 @@ turned on and off on the command line using `--nsfw_checker` and
At installation time, InvokeAI will ask whether the checker should be
activated by default (neither argument given on the command line). The
response is stored in the InvokeAI initialization file (usually
`invokeai.init` in your home directory). You can change the default at any
time by opening this file in a text editor and commenting or
uncommenting the line `--nsfw_checker`.
response is stored in the InvokeAI initialization file
(`invokeai.yaml` in the InvokeAI root directory). You can change the
default at any time by opening this file in a text editor and
changing the line `nsfw_checker:` from true to false or vice-versa:
```
...
Features:
esrgan: true
internet_available: true
log_tokenization: false
nsfw_checker: true
patchmatch: true
restore: true
```
## Caveats
@ -79,11 +91,3 @@ generates. However, it does write metadata into the PNG data area,
including the prompt used to generate the image and relevant parameter
settings. These fields can be examined using the `sd-metadata.py`
script that comes with the InvokeAI package.
Note that several other Stable Diffusion distributions offer
wavelet-based "invisible" watermarking. We have experimented with the
library used to generate these watermarks and have reached the
conclusion that while the watermarking library may be adding
watermarks to PNG images, the currently available version is unable to
retrieve them successfully. If and when a functioning version of the
library becomes available, we will offer this feature as well.

116
docs/features/OTHER.md
View File

@ -18,43 +18,16 @@ Output Example:
## **Seamless Tiling**
The seamless tiling mode causes generated images to seamlessly tile with itself. To use it, add the
`--seamless` option when starting the script which will result in all generated images to tile, or
for each `invoke>` prompt as shown here:
The seamless tiling mode causes generated images to seamlessly tile
with itself creating repetitive wallpaper-like patterns. To use it,
activate the Seamless Tiling option in the Web GUI and then select
whether to tile on the X (horizontal) and/or Y (vertical) axes. Tiling
will then be active for the next set of generations.
A nice prompt to test seamless tiling with is:
```python
invoke> "pond garden with lotus by claude monet" --seamless -s100 -n4
```
By default this will tile on both the X and Y axes. However, you can also specify specific axes to tile on with `--seamless_axes`.
Possible values are `x`, `y`, and `x,y`:
```python
invoke> "pond garden with lotus by claude monet" --seamless --seamless_axes=x -s100 -n4
```
---
## **Shortcuts: Reusing Seeds**
Since it is so common to reuse seeds while refining a prompt, there is now a shortcut as of version
1.11. Provide a `-S` (or `--seed`) switch of `-1` to use the seed of the most recent image
generated. If you produced multiple images with the `-n` switch, then you can go back further
using `-2`, `-3`, etc. up to the first image generated by the previous command. Sorry, but you can't go
back further than one command.
Here's an example of using this to do a quick refinement. It also illustrates using the new `-G`
switch to turn on upscaling and face enhancement (see previous section):
```bash
invoke> a cute child playing hopscotch -G0.5
[...]
outputs/img-samples/000039.3498014304.png: "a cute child playing hopscotch" -s50 -W512 -H512 -C7.5 -mk_lms -S3498014304
# I wonder what it will look like if I bump up the steps and set facial enhancement to full strength?
invoke> a cute child playing hopscotch -G1.0 -s100 -S -1
reusing previous seed 3498014304
[...]
outputs/img-samples/000040.3498014304.png: "a cute child playing hopscotch" -G1.0 -s100 -W512 -H512 -C7.5 -mk_lms -S3498014304
pond garden with lotus by claude monet"
```
---
@ -73,66 +46,27 @@ This will tell the sampler to invest 25% of its effort on the tabby cat aspect o
on the white duck aspect (surprisingly, this example actually works). The prompt weights can use any
combination of integers and floating point numbers, and they do not need to add up to 1.
---
## **Filename Format**
The argument `--fnformat` allows to specify the filename of the
image. Supported wildcards are all arguments what can be set such as
`perlin`, `seed`, `threshold`, `height`, `width`, `gfpgan_strength`,
`sampler_name`, `steps`, `model`, `upscale`, `prompt`, `cfg_scale`,
`prefix`.
The following prompt
```bash
dream> a red car --steps 25 -C 9.8 --perlin 0.1 --fnformat {prompt}_steps.{steps}_cfg.{cfg_scale}_perlin.{perlin}.png
```
generates a file with the name: `outputs/img-samples/a red car_steps.25_cfg.9.8_perlin.0.1.png`
---
## **Thresholding and Perlin Noise Initialization Options**
Two new options are the thresholding (`--threshold`) and the perlin noise initialization (`--perlin`) options. Thresholding limits the range of the latent values during optimization, which helps combat oversaturation with higher CFG scale values. Perlin noise initialization starts with a percentage (a value ranging from 0 to 1) of perlin noise mixed into the initial noise. Both features allow for more variations and options in the course of generating images.
Under the Noise section of the Web UI, you will find two options named
Perlin Noise and Noise Threshold. [Perlin
noise](https://en.wikipedia.org/wiki/Perlin_noise) is a type of
structured noise used to simulate terrain and other natural
textures. The slider controls the percentage of perlin noise that will
be mixed into the image at the beginning of generation. Adding a little
perlin noise to a generation will alter the image substantially.
The noise threshold limits the range of the latent values during
sampling and helps combat the oversharpening seem with higher CFG
scale values.
For better intuition into what these options do in practice:
![here is a graphic demonstrating them both](../assets/truncation_comparison.jpg)
In generating this graphic, perlin noise at initialization was programmatically varied going across on the diagram by values 0.0, 0.1, 0.2, 0.4, 0.5, 0.6, 0.8, 0.9, 1.0; and the threshold was varied going down from
0, 1, 2, 3, 4, 5, 10, 20, 100. The other options are fixed, so the initial prompt is as follows (no thresholding or perlin noise):
```bash
invoke> "a portrait of a beautiful young lady" -S 1950357039 -s 100 -C 20 -A k_euler_a --threshold 0 --perlin 0
```
Here's an example of another prompt used when setting the threshold to 5 and perlin noise to 0.2:
```bash
invoke> "a portrait of a beautiful young lady" -S 1950357039 -s 100 -C 20 -A k_euler_a --threshold 5 --perlin 0.2
```
!!! note
currently the thresholding feature is only implemented for the k-diffusion style samplers, and empirically appears to work best with `k_euler_a` and `k_dpm_2_a`. Using 0 disables thresholding. Using 0 for perlin noise disables using perlin noise for initialization. Finally, using 1 for perlin noise uses only perlin noise for initialization.
---
## **Simplified API**
For programmers who wish to incorporate stable-diffusion into other products, this repository
includes a simplified API for text to image generation, which lets you create images from a prompt
in just three lines of code:
```bash
from ldm.generate import Generate
g = Generate()
outputs = g.txt2img("a unicorn in manhattan")
```
Outputs is a list of lists in the format [filename1,seed1],[filename2,seed2]...].
Please see the documentation in ldm/generate.py for more information.
---
In generating this graphic, perlin noise at initialization was
programmatically varied going across on the diagram by values 0.0,
0.1, 0.2, 0.4, 0.5, 0.6, 0.8, 0.9, 1.0; and the threshold was varied
going down from 0, 1, 2, 3, 4, 5, 10, 20, 100. The other options are
fixed using the prompt "a portrait of a beautiful young lady" a CFG of
20, 100 steps, and a seed of 1950357039.

161
docs/features/POSTPROCESS.md
View File

@ -8,12 +8,6 @@ title: Postprocessing
This extension provides the ability to restore faces and upscale images.
Face restoration and upscaling can be applied at the time you generate the
images, or at any later time against a previously-generated PNG file, using the
[!fix](#fixing-previously-generated-images) command.
[Outpainting and outcropping](OUTPAINTING.md) can only be applied after the
fact.
## Face Fixing
The default face restoration module is GFPGAN. The default upscale is
@ -23,8 +17,7 @@ Real-ESRGAN. For an alternative face restoration module, see
As of version 1.14, environment.yaml will install the Real-ESRGAN package into
the standard install location for python packages, and will put GFPGAN into a
subdirectory of "src" in the InvokeAI directory. Upscaling with Real-ESRGAN
should "just work" without further intervention. Simply pass the `--upscale`
(`-U`) option on the `invoke>` command line, or indicate the desired scale on
should "just work" without further intervention. Simply indicate the desired scale on
the popup in the Web GUI.
**GFPGAN** requires a series of downloadable model files to work. These are
@ -41,48 +34,75 @@ reconstruction.
### Upscaling
`-U : <upscaling_factor> <upscaling_strength>`
Open the upscaling dialog by clicking on the "expand" icon located
above the image display area in the Web UI:
The upscaling prompt argument takes two values. The first value is a scaling
factor and should be set to either `2` or `4` only. This will either scale the
image 2x or 4x respectively using different models.
<figure markdown>
![upscale1](../assets/features/upscale-dialog.png)
</figure>
You can set the scaling stength between `0` and `1.0` to control intensity of
the of the scaling. This is handy because AI upscalers generally tend to smooth
out texture details. If you wish to retain some of those for natural looking
results, we recommend using values between `0.5 to 0.8`.
There are three different upscaling parameters that you can
adjust. The first is the scale itself, either 2x or 4x.
If you do not explicitly specify an upscaling_strength, it will default to 0.75.
The second is the "Denoising Strength." Higher values will smooth out
the image and remove digital chatter, but may lose fine detail at
higher values.
Third, "Upscale Strength" allows you to adjust how the You can set the
scaling stength between `0` and `1.0` to control the intensity of the
scaling. AI upscalers generally tend to smooth out texture details. If
you wish to retain some of those for natural looking results, we
recommend using values between `0.5 to 0.8`.
[This figure](../assets/features/upscaling-montage.png) illustrates
the effects of denoising and strength. The original image was 512x512,
4x scaled to 2048x2048. The "original" version on the upper left was
scaled using simple pixel averaging. The remainder use the ESRGAN
upscaling algorithm at different levels of denoising and strength.
<figure markdown>
![upscaling](../assets/features/upscaling-montage.png){ width=720 }
</figure>
Both denoising and strength default to 0.75.
### Face Restoration
`-G : <facetool_strength>`
InvokeAI offers alternative two face restoration algorithms,
[GFPGAN](https://github.com/TencentARC/GFPGAN) and
[CodeFormer](https://huggingface.co/spaces/sczhou/CodeFormer). These
algorithms improve the appearance of faces, particularly eyes and
mouths. Issues with faces are less common with the latest set of
Stable Diffusion models than with the original 1.4 release, but the
restoration algorithms can still make a noticeable improvement in
certain cases. You can also apply restoration to old photographs you
upload.
This prompt argument controls the strength of the face restoration that is being
applied. Similar to upscaling, values between `0.5 to 0.8` are recommended.
To access face restoration, click the "smiley face" icon in the
toolbar above the InvokeAI image panel. You will be presented with a
dialog that offers a choice between the two algorithm and sliders that
allow you to adjust their parameters. Alternatively, you may open the
left-hand accordion panel labeled "Face Restoration" and have the
restoration algorithm of your choice applied to generated images
automatically.
You can use either one or both without any conflicts. In cases where you use
both, the image will be first upscaled and then the face restoration process
will be executed to ensure you get the highest quality facial features.
`--save_orig`
Like upscaling, there are a number of parameters that adjust the face
restoration output. GFPGAN has a single parameter, `strength`, which
controls how much the algorithm is allowed to adjust the
image. CodeFormer has two parameters, `strength`, and `fidelity`,
which together control the quality of the output image as described in
the [CodeFormer project
page](https://shangchenzhou.com/projects/CodeFormer/). Default values
are 0.75 for both parameters, which achieves a reasonable balance
between changing the image too much and not enough.
When you use either `-U` or `-G`, the final result you get is upscaled or face
modified. If you want to save the original Stable Diffusion generation, you can
use the `-save_orig` prompt argument to save the original unaffected version
too.
[This figure](../assets/features/restoration-montage.png) illustrates
the effects of adjusting GFPGAN and CodeFormer parameters.
### Example Usage
```bash
invoke> "superman dancing with a panda bear" -U 2 0.6 -G 0.4
```
This also works with img2img:
```bash
invoke> "a man wearing a pineapple hat" -I path/to/your/file.png -U 2 0.5 -G 0.6
```
<figure markdown>
![upscaling](../assets/features/restoration-montage.png){ width=720 }
</figure>
!!! note
@ -95,69 +115,8 @@ invoke> "a man wearing a pineapple hat" -I path/to/your/file.png -U 2 0.5 -G 0.6
process is complete. While the image generation is taking place, you will still be able to preview
the base images.
If you wish to stop during the image generation but want to upscale or face
restore a particular generated image, pass it again with the same prompt and
generated seed along with the `-U` and `-G` prompt arguments to perform those
actions.
## CodeFormer Support
This repo also allows you to perform face restoration using
[CodeFormer](https://github.com/sczhou/CodeFormer).
In order to setup CodeFormer to work, you need to download the models like with
GFPGAN. You can do this either by running `invokeai-configure` or by manually
downloading the
[model file](https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/codeformer.pth)
and saving it to `ldm/invoke/restoration/codeformer/weights` folder.
You can use `-ft` prompt argument to swap between CodeFormer and the default
GFPGAN. The above mentioned `-G` prompt argument will allow you to control the
strength of the restoration effect.
### CodeFormer Usage
The following command will perform face restoration with CodeFormer instead of
the default gfpgan.
`<prompt> -G 0.8 -ft codeformer`
### Other Options
- `-cf` - cf or CodeFormer Fidelity takes values between `0` and `1`. 0 produces
high quality results but low accuracy and 1 produces lower quality results but
higher accuacy to your original face.
The following command will perform face restoration with CodeFormer. CodeFormer
will output a result that is closely matching to the input face.
`<prompt> -G 1.0 -ft codeformer -cf 0.9`
The following command will perform face restoration with CodeFormer. CodeFormer
will output a result that is the best restoration possible. This may deviate
slightly from the original face. This is an excellent option to use in
situations when there is very little facial data to work with.
`<prompt> -G 1.0 -ft codeformer -cf 0.1`
## Fixing Previously-Generated Images
It is easy to apply face restoration and/or upscaling to any
previously-generated file. Just use the syntax
`!fix path/to/file.png <options>`. For example, to apply GFPGAN at strength 0.8
and upscale 2X for a file named `./outputs/img-samples/000044.2945021133.png`,
just run:
```bash
invoke> !fix ./outputs/img-samples/000044.2945021133.png -G 0.8 -U 2
```
A new file named `000044.2945021133.fixed.png` will be created in the output
directory. Note that the `!fix` command does not replace the original file,
unlike the behavior at generate time.
## How to disable
If, for some reason, you do not wish to load the GFPGAN and/or ESRGAN libraries,
you can disable them on the invoke.py command line with the `--no_restore` and
`--no_upscale` options, respectively.
`--no_esrgan` options, respectively.

147
docs/features/PROMPTS.md
View File

@ -4,77 +4,12 @@ title: Prompting-Features
# :octicons-command-palette-24: Prompting-Features
## **Reading Prompts from a File**
You can automate `invoke.py` by providing a text file with the prompts you want
to run, one line per prompt. The text file must be composed with a text editor
(e.g. Notepad) and not a word processor. Each line should look like what you
would type at the invoke> prompt:
```bash
"a beautiful sunny day in the park, children playing" -n4 -C10
"stormy weather on a mountain top, goats grazing" -s100
"innovative packaging for a squid's dinner" -S137038382
```
Then pass this file's name to `invoke.py` when you invoke it:
```bash
python scripts/invoke.py --from_file "/path/to/prompts.txt"
```
You may also read a series of prompts from standard input by providing
a filename of `-`. For example, here is a python script that creates a
matrix of prompts, each one varying slightly:
```bash
#!/usr/bin/env python
adjectives = ['sunny','rainy','overcast']
samplers = ['k_lms','k_euler_a','k_heun']
cfg = [7.5, 9, 11]
for adj in adjectives:
for samp in samplers:
for cg in cfg:
print(f'a {adj} day -A{samp} -C{cg}')
```
Its output looks like this (abbreviated):
```bash
a sunny day -Aklms -C7.5
a sunny day -Aklms -C9
a sunny day -Aklms -C11
a sunny day -Ak_euler_a -C7.5
a sunny day -Ak_euler_a -C9
...
a overcast day -Ak_heun -C9
a overcast day -Ak_heun -C11
```
To feed it to invoke.py, pass the filename of "-"
```bash
python matrix.py | python scripts/invoke.py --from_file -
```
When the script is finished, each of the 27 combinations
of adjective, sampler and CFG will be executed.
The command-line interface provides `!fetch` and `!replay` commands
which allow you to read the prompts from a single previously-generated
image or a whole directory of them, write the prompts to a file, and
then replay them. Or you can create your own file of prompts and feed
them to the command-line client from within an interactive session.
See [Command-Line Interface](CLI.md) for details.
---
## **Negative and Unconditioned Prompts**
Any words between a pair of square brackets will instruct Stable Diffusion to
attempt to ban the concept from the generated image.
Any words between a pair of square brackets will instruct Stable
Diffusion to attempt to ban the concept from the generated image. The
same effect is achieved by placing words in the "Negative Prompts"
textbox in the Web UI.
```text
this is a test prompt [not really] to make you understand [cool] how this works.
@ -87,7 +22,9 @@ Here's a prompt that depicts what it does.
original prompt:
`#!bash "A fantastical translucent pony made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve" -s 20 -W 512 -H 768 -C 7.5 -A k_euler_a -S 1654590180`
`#!bash "A fantastical translucent pony made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve"`
`#!bash parameters: steps=20, dimensions=512x768, CFG=7.5, Scheduler=k_euler_a, seed=1654590180`
<figure markdown>
@ -99,7 +36,8 @@ That image has a woman, so if we want the horse without a rider, we can
influence the image not to have a woman by putting [woman] in the prompt, like
this:
`#!bash "A fantastical translucent poney made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve [woman]" -s 20 -W 512 -H 768 -C 7.5 -A k_euler_a -S 1654590180`
`#!bash "A fantastical translucent poney made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve [woman]"`
(same parameters as above)
<figure markdown>
@ -110,7 +48,8 @@ this:
That's nice - but say we also don't want the image to be quite so blue. We can
add "blue" to the list of negative prompts, so it's now [woman blue]:
`#!bash "A fantastical translucent poney made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve [woman blue]" -s 20 -W 512 -H 768 -C 7.5 -A k_euler_a -S 1654590180`
`#!bash "A fantastical translucent poney made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve [woman blue]"`
(same parameters as above)
<figure markdown>
@ -121,7 +60,8 @@ add "blue" to the list of negative prompts, so it's now [woman blue]:
Getting close - but there's no sense in having a saddle when our horse doesn't
have a rider, so we'll add one more negative prompt: [woman blue saddle].
`#!bash "A fantastical translucent poney made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve [woman blue saddle]" -s 20 -W 512 -H 768 -C 7.5 -A k_euler_a -S 1654590180`
`#!bash "A fantastical translucent poney made of water and foam, ethereal, radiant, hyperalism, scottish folklore, digital painting, artstation, concept art, smooth, 8 k frostbite 3 engine, ultra detailed, art by artgerm and greg rutkowski and magali villeneuve [woman blue saddle]"`
(same parameters as above)
<figure markdown>
@ -261,19 +201,6 @@ Prompt2prompt `.swap()` is not compatible with xformers, which will be temporari
The `prompt2prompt` code is based off
[bloc97's colab](https://github.com/bloc97/CrossAttentionControl).
Note that `prompt2prompt` is not currently working with the runwayML inpainting
model, and may never work due to the way this model is set up. If you attempt to
use `prompt2prompt` you will get the original image back. However, since this
model is so good at inpainting, a good substitute is to use the `clipseg` text
masking option:
```bash
invoke> a fluffy cat eating a hotdog
Outputs:
[1010] outputs/000025.2182095108.png: a fluffy cat eating a hotdog
invoke> a smiling dog eating a hotdog -I 000025.2182095108.png -tm cat
```
### Escaping parantheses () and speech marks ""
If the model you are using has parentheses () or speech marks "" as part of its
@ -374,6 +301,48 @@ summoning up the concept of some sort of scifi creature? Let's find out.
Indeed, removing the word "hybrid" produces an image that is more like what we'd
expect.
In conclusion, prompt blending is great for exploring creative space, but can be
difficult to direct. A forthcoming release of InvokeAI will feature more
deterministic prompt weighting.
## Dynamic Prompts
Dynamic Prompts are a powerful feature designed to produce a variety of prompts based on user-defined options. Using a special syntax, you can construct a prompt with multiple possibilities, and the system will automatically generate a series of permutations based on your settings. This is extremely beneficial for ideation, exploring various scenarios, or testing different concepts swiftly and efficiently.
### Structure of a Dynamic Prompt
A Dynamic Prompt comprises of regular text, supplemented with alternatives enclosed within curly braces {} and separated by a vertical bar |. For example: {option1|option2|option3}. The system will then select one of the options to include in the final prompt. This flexible system allows for options to be placed throughout the text as needed.
Furthermore, Dynamic Prompts can designate multiple selections from a single group of options. This feature is triggered by prefixing the options with a numerical value followed by $$. For example, in {2$$option1|option2|option3}, the system will select two distinct options from the set.
### Creating Dynamic Prompts
To create a Dynamic Prompt, follow these steps:
Draft your sentence or phrase, identifying words or phrases with multiple possible options.
Encapsulate the different options within curly braces {}.
Within the braces, separate each option using a vertical bar |.
If you want to include multiple options from a single group, prefix with the desired number and $$.
For instance: A {house|apartment|lodge|cottage} in {summer|winter|autumn|spring} designed in {2$$style1|style2|style3}.
### How Dynamic Prompts Work
Once a Dynamic Prompt is configured, the system generates an array of combinations using the options provided. Each group of options in curly braces is treated independently, with the system selecting one option from each group. For a prefixed set (e.g., 2$$), the system will select two distinct options.
For example, the following prompts could be generated from the above Dynamic Prompt:
A house in summer designed in style1, style2
A lodge in autumn designed in style3, style1
A cottage in winter designed in style2, style3
And many more!
When the `Combinatorial` setting is on, Invoke will disable the "Images" selection, and generate every combination up until the setting for Max Prompts is reached.
When the `Combinatorial` setting is off, Invoke will randomly generate combinations up until the setting for Images has been reached.
### Tips and Tricks for Using Dynamic Prompts
Below are some useful strategies for creating Dynamic Prompts:
Utilize Dynamic Prompts to generate a wide spectrum of prompts, perfect for brainstorming and exploring diverse ideas.
Ensure that the options within a group are contextually relevant to the part of the sentence where they are used. For instance, group building types together, and seasons together.
Apply the 2$$ prefix when you want to incorporate more than one option from a single group. This becomes quite handy when mixing and matching different elements.
Experiment with different quantities for the prefix. For example, 3$$ will select three distinct options.
Be aware of coherence in your prompts. Although the system can generate all possible combinations, not all may semantically make sense. Therefore, carefully choose the options for each group.
Always review and fine-tune the generated prompts as needed. While Dynamic Prompts can help you generate a multitude of combinations, the final polishing and refining remain in your hands.

33
docs/features/TEXTUAL_INVERSION.md → docs/features/TRAINING.md
View File

@ -1,9 +1,10 @@
---
title: Textual-Inversion
title: Training
---
# :material-file-document: Textual Inversion
# :material-file-document: Training
# Textual Inversion Training
## **Personalizing Text-to-Image Generation**
You may personalize the generated images to provide your own styles or objects
@ -46,11 +47,19 @@ start the front end by selecting choice (3):
```sh
Do you want to generate images using the
1. command-line
2. browser-based UI
3. textual inversion training
4. open the developer console
Please enter 1, 2, 3, or 4: [1] 3
1: Browser-based UI
2: Command-line interface
3: Run textual inversion training
4: Merge models (diffusers type only)
5: Download and install models
6: Change InvokeAI startup options
7: Re-run the configure script to fix a broken install
8: Open the developer console
9: Update InvokeAI
10: Command-line help
Q: Quit
Please enter 1-10, Q: [1]
```
From the command line, with the InvokeAI virtual environment active,
@ -250,16 +259,6 @@ invokeai-ti \
--only_save_embeds
```
## Using Embeddings
After training completes, the resultant embeddings will be saved into your `$INVOKEAI_ROOT/embeddings/<trigger word>/learned_embeds.bin`.
These will be automatically loaded when you start InvokeAI.
Add the trigger word, surrounded by angle brackets, to use that embedding. For example, if your trigger word was `terence`, use `<terence>` in prompts. This is the same syntax used by the HuggingFace concepts library.
**Note:** `.pt` embeddings do not require the angle brackets.
## Troubleshooting
### `Cannot load embedding for <trigger>. It was trained on a model with token dimension 1024, but the current model has token dimension 768`

42
docs/features/VARIATIONS.md
View File

@ -6,9 +6,7 @@ title: Variations
## Intro
Release 1.13 of SD-Dream adds support for image variations.
You are able to do the following:
InvokeAI's support for variations enables you to do the following:
1. Generate a series of systematic variations of an image, given a prompt. The
amount of variation from one image to the next can be controlled.
@ -30,19 +28,7 @@ The prompt we will use throughout is:
This will be indicated as `#!bash "prompt"` in the examples below.
First we let SD create a series of images in the usual way, in this case
requesting six iterations:
```bash
invoke> lucy lawless as xena, warrior princess, character portrait, high resolution -n6
...
Outputs:
./outputs/Xena/000001.1579445059.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -S1579445059
./outputs/Xena/000001.1880768722.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -S1880768722
./outputs/Xena/000001.332057179.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -S332057179
./outputs/Xena/000001.2224800325.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -S2224800325
./outputs/Xena/000001.465250761.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -S465250761
./outputs/Xena/000001.3357757885.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -S3357757885
```
requesting six iterations.
<figure markdown>
![var1](../assets/variation_walkthru/000001.3357757885.png)
@ -53,22 +39,16 @@ Outputs:
## Step 2 - Generating Variations
Let's try to generate some variations. Using the same seed, we pass the argument
`-v0.1` (or --variant_amount), which generates a series of variations each
differing by a variation amount of 0.2. This number ranges from `0` to `1.0`,
with higher numbers being larger amounts of variation.
Let's try to generate some variations on this image. We select the "*"
symbol in the line of icons above the image in order to fix the prompt
and seed. Then we open up the "Variations" section of the generation
panel and use the slider to set the variation amount to 0.2. The
higher this value, the more each generated image will differ from the
previous one.
```bash
invoke> "prompt" -n6 -S3357757885 -v0.2
...
Outputs:
./outputs/Xena/000002.784039624.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -V 784039624:0.2 -S3357757885
./outputs/Xena/000002.3647897225.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -V 3647897225:0.2 -S3357757885
./outputs/Xena/000002.917731034.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -V 917731034:0.2 -S3357757885
./outputs/Xena/000002.4116285959.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -V 4116285959:0.2 -S3357757885
./outputs/Xena/000002.1614299449.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -V 1614299449:0.2 -S3357757885
./outputs/Xena/000002.1335553075.png: "prompt" -s50 -W512 -H512 -C7.5 -Ak_lms -V 1335553075:0.2 -S3357757885
```
Now we run the prompt a second time, requesting six iterations. You
will see six images that are thematically related to each other. Try
increasing and decreasing the variation amount and see what happens.
### **Variation Sub Seeding**

10
docs/features/WEB.md
View File

@ -299,14 +299,6 @@ initial image" icons are located.
See the [Unified Canvas Guide](UNIFIED_CANVAS.md)
## Parting remarks
This concludes the walkthrough, but there are several more features that you can
explore. Please check out the [Command Line Interface](CLI.md) documentation for
further explanation of the advanced features that were not covered here.
The WebUI is only rapid development. Check back regularly for updates!
## Reference
### Additional Options
@ -349,11 +341,9 @@ the settings configured in the toolbar.
See below for additional documentation related to each feature:
- [Core Prompt Settings](./CLI.md)
- [Variations](./VARIATIONS.md)
- [Upscaling](./POSTPROCESS.md#upscaling)
- [Image to Image](./IMG2IMG.md)
- [Inpainting](./INPAINTING.md)
- [Other](./OTHER.md)
#### Invocation Gallery

14
docs/features/index.md
View File

@ -13,28 +13,16 @@ Build complex scenes by combine and modifying multiple images in a stepwise
fashion. This feature combines img2img, inpainting and outpainting in
a single convenient digital artist-optimized user interface.
### * The [Command Line Interface (CLI)](CLI.md)
Scriptable access to InvokeAI's features.
## Image Generation
### * [Prompt Engineering](PROMPTS.md)
Get the images you want with the InvokeAI prompt engineering language.
## * [Post-Processing](POSTPROCESS.md)
Restore mangled faces and make images larger with upscaling. Also see the [Embiggen Upscaling Guide](EMBIGGEN.md).
## * The [Concepts Library](CONCEPTS.md)
Add custom subjects and styles using HuggingFace's repository of embeddings.
### * [Image-to-Image Guide for the CLI](IMG2IMG.md)
### * [Image-to-Image Guide](IMG2IMG.md)
Use a seed image to build new creations in the CLI.
### * [Inpainting Guide for the CLI](INPAINTING.md)
Selectively erase and replace portions of an existing image in the CLI.
### * [Outpainting Guide for the CLI](OUTPAINTING.md)
Extend the borders of the image with an "outcrop" function within the CLI.
### * [Generating Variations](VARIATIONS.md)
Have an image you like and want to generate many more like it? Variations
are the ticket.

122
docs/index.md
View File

@ -13,6 +13,7 @@ title: Home
<div align="center" markdown>
[![project logo](assets/invoke_ai_banner.png)](https://github.com/invoke-ai/InvokeAI)
[![discord badge]][discord link]
@ -131,17 +132,13 @@ This method is recommended for those familiar with running Docker containers
- [WebUI overview](features/WEB.md)
- [WebUI hotkey reference guide](features/WEBUIHOTKEYS.md)
- [WebUI Unified Canvas for Img2Img, inpainting and outpainting](features/UNIFIED_CANVAS.md)
<!-- separator -->
### The InvokeAI Command Line Interface
- [Command Line Interace Reference Guide](features/CLI.md)
<!-- separator -->
### Image Management
- [Image2Image](features/IMG2IMG.md)
- [Inpainting](features/INPAINTING.md)
- [Outpainting](features/OUTPAINTING.md)
- [Adding custom styles and subjects](features/CONCEPTS.md)
- [Upscaling and Face Reconstruction](features/POSTPROCESS.md)
- [Embiggen upscaling](features/EMBIGGEN.md)
- [Other Features](features/OTHER.md)
<!-- separator -->
@ -156,83 +153,63 @@ This method is recommended for those familiar with running Docker containers
- [Prompt Syntax](features/PROMPTS.md)
- [Generating Variations](features/VARIATIONS.md)
## :octicons-log-16: Latest Changes
### InvokeAI Configuration
- [Guide to InvokeAI Runtime Settings](features/CONFIGURATION.md)
### v2.3.0 <small>(9 February 2023)</small>
## :octicons-log-16: Important Changes Since Version 2.3
#### Migration to Stable Diffusion `diffusers` models
### Nodes
Previous versions of InvokeAI supported the original model file format introduced with Stable Diffusion 1.4. In the original format, known variously as "checkpoint", or "legacy" format, there is a single large weights file ending with `.ckpt` or `.safetensors`. Though this format has served the community well, it has a number of disadvantages, including file size, slow loading times, and a variety of non-standard variants that require special-case code to handle. In addition, because checkpoint files are actually a bundle of multiple machine learning sub-models, it is hard to swap different sub-models in and out, or to share common sub-models. A new format, introduced by the StabilityAI company in collaboration with HuggingFace, is called `diffusers` and consists of a directory of individual models. The most immediate benefit of `diffusers` is that they load from disk very quickly. A longer term benefit is that in the near future `diffusers` models will be able to share common sub-models, dramatically reducing disk space when you have multiple fine-tune models derived from the same base.
Behind the scenes, InvokeAI has been completely rewritten to support
"nodes," small unitary operations that can be combined into graphs to
form arbitrary workflows. For example, there is a prompt node that
processes the prompt string and feeds it to a text2latent node that
generates a latent image. The latents are then fed to a latent2image
node that translates the latent image into a PNG.
When you perform a new install of version 2.3.0, you will be offered the option to install the `diffusers` versions of a number of popular SD models, including Stable Diffusion versions 1.5 and 2.1 (including the 768x768 pixel version of 2.1). These will act and work just like the checkpoint versions. Do not be concerned if you already have a lot of ".ckpt" or ".safetensors" models on disk! InvokeAI 2.3.0 can still load these and generate images from them without any extra intervention on your part.
The WebGUI has a node editor that allows you to graphically design and
execute custom node graphs. The ability to save and load graphs is
still a work in progress, but coming soon.
To take advantage of the optimized loading times of `diffusers` models, InvokeAI offers options to convert legacy checkpoint models into optimized `diffusers` models. If you use the `invokeai` command line interface, the relevant commands are:
### Command-Line Interface Retired
* `!convert_model` -- Take the path to a local checkpoint file or a URL that is pointing to one, convert it into a `diffusers` model, and import it into InvokeAI's models registry file.
* `!optimize_model` -- If you already have a checkpoint model in your InvokeAI models file, this command will accept its short name and convert it into a like-named `diffusers` model, optionally deleting the original checkpoint file.
* `!import_model` -- Take the local path of either a checkpoint file or a `diffusers` model directory and import it into InvokeAI's registry file. You may also provide the ID of any diffusers model that has been published on the [HuggingFace models repository](https://huggingface.co/models?pipeline_tag=text-to-image&sort=downloads) and it will be downloaded and installed automatically.
The original "invokeai" command-line interface has been retired. The
`invokeai` command will now launch a new command-line client that can
be used by developers to create and test nodes. It is not intended to
be used for routine image generation or manipulation.
The WebGUI offers similar functionality for model management.
To launch the Web GUI from the command-line, use the command
`invokeai-web` rather than the traditional `invokeai --web`.
For advanced users, new command-line options provide additional functionality. Launching `invokeai` with the argument `--autoconvert <path to directory>` takes the path to a directory of checkpoint files, automatically converts them into `diffusers` models and imports them. Each time the script is launched, the directory will be scanned for new checkpoint files to be loaded. Alternatively, the `--ckpt_convert` argument will cause any checkpoint or safetensors model that is already registered with InvokeAI to be converted into a `diffusers` model on the fly, allowing you to take advantage of future diffusers-only features without explicitly converting the model and saving it to disk.
### ControlNet
Please see [INSTALLING MODELS](https://invoke-ai.github.io/InvokeAI/installation/050_INSTALLING_MODELS/) for more information on model management in both the command-line and Web interfaces.
This version of InvokeAI features ControlNet, a system that allows you
to achieve exact poses for human and animal figures by providing a
model to follow. Full details are found in [ControlNet](features/CONTROLNET.md)
#### Support for the `XFormers` Memory-Efficient Crossattention Package
### New Schedulers
On CUDA (Nvidia) systems, version 2.3.0 supports the `XFormers` library. Once installed, the`xformers` package dramatically reduces the memory footprint of loaded Stable Diffusion models files and modestly increases image generation speed. `xformers` will be installed and activated automatically if you specify a CUDA system at install time.
The list of schedulers has been completely revamped and brought up to date:
The caveat with using `xformers` is that it introduces slightly non-deterministic behavior, and images generated using the same seed and other settings will be subtly different between invocations. Generally the changes are unnoticeable unless you rapidly shift back and forth between images, but to disable `xformers` and restore fully deterministic behavior, you may launch InvokeAI using the `--no-xformers` option. This is most conveniently done by opening the file `invokeai/invokeai.init` with a text editor, and adding the line `--no-xformers` at the bottom.
| **Short Name** | **Scheduler** | **Notes** |
|----------------|---------------------------------|-----------------------------|
| **ddim** | DDIMScheduler | |
| **ddpm** | DDPMScheduler | |
| **deis** | DEISMultistepScheduler | |
| **lms** | LMSDiscreteScheduler | |
| **pndm** | PNDMScheduler | |
| **heun** | HeunDiscreteScheduler | original noise schedule |
| **heun_k** | HeunDiscreteScheduler | using karras noise schedule |
| **euler** | EulerDiscreteScheduler | original noise schedule |
| **euler_k** | EulerDiscreteScheduler | using karras noise schedule |
| **kdpm_2** | KDPM2DiscreteScheduler | |
| **kdpm_2_a** | KDPM2AncestralDiscreteScheduler | |
| **dpmpp_2s** | DPMSolverSinglestepScheduler | |
| **dpmpp_2m** | DPMSolverMultistepScheduler | original noise scnedule |
| **dpmpp_2m_k** | DPMSolverMultistepScheduler | using karras noise schedule |
| **unipc** | UniPCMultistepScheduler | CPU only |
#### A Negative Prompt Box in the WebUI
There is now a separate text input box for negative prompts in the WebUI. This is convenient for stashing frequently-used negative prompts ("mangled limbs, bad anatomy"). The `[negative prompt]` syntax continues to work in the main prompt box as well.
To see exactly how your prompts are being parsed, launch `invokeai` with the `--log_tokenization` option. The console window will then display the tokenization process for both positive and negative prompts.
#### Model Merging
Version 2.3.0 offers an intuitive user interface for merging up to three Stable Diffusion models using an intuitive user interface. Model merging allows you to mix the behavior of models to achieve very interesting effects. To use this, each of the models must already be imported into InvokeAI and saved in `diffusers` format, then launch the merger using a new menu item in the InvokeAI launcher script (`invoke.sh`, `invoke.bat`) or directly from the command line with `invokeai-merge --gui`. You will be prompted to select the models to merge, the proportions in which to mix them, and the mixing algorithm. The script will create a new merged `diffusers` model and import it into InvokeAI for your use.
See [MODEL MERGING](https://invoke-ai.github.io/InvokeAI/features/MODEL_MERGING/) for more details.
#### Textual Inversion Training
Textual Inversion (TI) is a technique for training a Stable Diffusion model to emit a particular subject or style when triggered by a keyword phrase. You can perform TI training by placing a small number of images of the subject or style in a directory, and choosing a distinctive trigger phrase, such as "pointillist-style". After successful training, The subject or style will be activated by including `<pointillist-style>` in your prompt.
Previous versions of InvokeAI were able to perform TI, but it required using a command-line script with dozens of obscure command-line arguments. Version 2.3.0 features an intuitive TI frontend that will build a TI model on top of any `diffusers` model. To access training you can launch from a new item in the launcher script or from the command line using `invokeai-ti --gui`.
See [TEXTUAL INVERSION](https://invoke-ai.github.io/InvokeAI/features/TEXTUAL_INVERSION/) for further details.
#### A New Installer Experience
The InvokeAI installer has been upgraded in order to provide a smoother and hopefully more glitch-free experience. In addition, InvokeAI is now packaged as a PyPi project, allowing developers and power-users to install InvokeAI with the command `pip install InvokeAI --use-pep517`. Please see [Installation](#installation) for details.
Developers should be aware that the `pip` installation procedure has been simplified and that the `conda` method is no longer supported at all. Accordingly, the `environments_and_requirements` directory has been deleted from the repository.
#### Command-line name changes
All of InvokeAI's functionality, including the WebUI, command-line interface, textual inversion training and model merging, can all be accessed from the `invoke.sh` and `invoke.bat` launcher scripts. The menu of options has been expanded to add the new functionality. For the convenience of developers and power users, we have normalized the names of the InvokeAI command-line scripts:
* `invokeai` -- Command-line client
* `invokeai --web` -- Web GUI
* `invokeai-merge --gui` -- Model merging script with graphical front end
* `invokeai-ti --gui` -- Textual inversion script with graphical front end
* `invokeai-configure` -- Configuration tool for initializing the `invokeai` directory and selecting popular starter models.
For backward compatibility, the old command names are also recognized, including `invoke.py` and `configure-invokeai.py`. However, these are deprecated and will eventually be removed.
Developers should be aware that the locations of the script's source code has been moved. The new locations are:
* `invokeai` => `ldm/invoke/CLI.py`
* `invokeai-configure` => `ldm/invoke/config/configure_invokeai.py`
* `invokeai-ti`=> `ldm/invoke/training/textual_inversion.py`
* `invokeai-merge` => `ldm/invoke/merge_diffusers`
Developers are strongly encouraged to perform an "editable" install of InvokeAI using `pip install -e . --use-pep517` in the Git repository, and then to call the scripts using their 2.3.0 names, rather than executing the scripts directly. Developers should also be aware that the several important data files have been relocated into a new directory named `invokeai`. This includes the WebGUI's `frontend` and `backend` directories, and the `INITIAL_MODELS.yaml` files used by the installer to select starter models. Eventually all InvokeAI modules will be in subdirectories of `invokeai`.
Please see [2.3.0 Release Notes](https://github.com/invoke-ai/InvokeAI/releases/tag/v2.3.0) for further details.
For older changelogs, please visit the
**[CHANGELOG](CHANGELOG/#v223-2-december-2022)**.
Please see [3.0.0 Release Notes](https://github.com/invoke-ai/InvokeAI/releases/tag/v3.0.0) for further details.
## :material-target: Troubleshooting
@ -268,8 +245,3 @@ free to send me an email if you use and like the script.
Original portions of the software are Copyright (c) 2022-23
by [The InvokeAI Team](https://github.com/invoke-ai).
## :octicons-book-24: Further Reading
Please see the original README for more information on this software and
underlying algorithm, located in the file
[README-CompViz.md](other/README-CompViz.md).

3
installer/lib/installer.py
View File

@ -149,7 +149,7 @@ class Installer:
return venv_dir
def install(self, root: str = "~/invokeai", version: str = "latest", yes_to_all=False, find_links: Path = None) -> None:
def install(self, root: str = "~/invokeai-3", version: str = "latest", yes_to_all=False, find_links: Path = None) -> None:
"""
Install the InvokeAI application into the given runtime path
@ -248,6 +248,7 @@ class InvokeAiInstance:
"install",
"--require-virtualenv",
"torch~=2.0.0",
"torchmetrics==0.11.4",
"torchvision>=0.14.1",
"--force-reinstall",
"--find-links" if find_links is not None else None,

6
installer/templates/invoke.bat.in
View File

@ -14,13 +14,13 @@ echo 3. Run textual inversion training
echo 4. Merge models (diffusers type only)
echo 5. Download and install models
echo 6. Change InvokeAI startup options
echo 7. Re-run the configure script to fix a broken install
echo 7. Re-run the configure script to fix a broken install or to complete a major upgrade
echo 8. Open the developer console
echo 9. Update InvokeAI
echo 10. Command-line help
echo Q - Quit
set /P choice="Please enter 1-10, Q: [2] "
if not defined choice set choice=2
if not defined choice set choice=1
IF /I "%choice%" == "1" (
echo Starting the InvokeAI browser-based UI..
python .venv\Scripts\invokeai-web.exe %*
@ -56,7 +56,7 @@ IF /I "%choice%" == "1" (
call cmd /k
) ELSE IF /I "%choice%" == "9" (
echo Running invokeai-update...
python .venv\Scripts\invokeai-update.exe %*
python -m invokeai.frontend.install.invokeai_update
) ELSE IF /I "%choice%" == "10" (
echo Displaying command line help...
python .venv\Scripts\invokeai.exe --help %*

8
installer/templates/invoke.sh.in
View File

@ -81,7 +81,7 @@ do_choice() {
;;
7)
clear
printf "Re-run the configure script to fix a broken install\n"
printf "Re-run the configure script to fix a broken install or to complete a major upgrade\n"
invokeai-configure --root ${INVOKEAI_ROOT} --yes --default_only
;;
8)
@ -93,7 +93,7 @@ do_choice() {
9)
clear
printf "Update InvokeAI\n"
invokeai-update
python -m invokeai.frontend.install.invokeai_update
;;
10)
clear
@ -118,12 +118,12 @@ do_choice() {
do_dialog() {
options=(
1 "Generate images with a browser-based interface"
2 "Generate images using a command-line interface"
2 "Explore InvokeAI nodes using a command-line interface"
3 "Textual inversion training"
4 "Merge models (diffusers type only)"
5 "Download and install models"
6 "Change InvokeAI startup options"
7 "Re-run the configure script to fix a broken install"
7 "Re-run the configure script to fix a broken install or to complete a major upgrade"
8 "Open the developer console"
9 "Update InvokeAI")

14
invokeai/app/api/dependencies.py
View File

@ -11,16 +11,16 @@ from invokeai.app.services.board_images import (
)
from invokeai.app.services.board_record_storage import SqliteBoardRecordStorage
from invokeai.app.services.boards import BoardService, BoardServiceDependencies
from invokeai.app.services.config import InvokeAIAppConfig
from invokeai.app.services.image_record_storage import SqliteImageRecordStorage
from invokeai.app.services.images import ImageService, ImageServiceDependencies
from invokeai.app.services.metadata import CoreMetadataService
from invokeai.app.services.resource_name import SimpleNameService
from invokeai.app.services.urls import LocalUrlService
from invokeai.backend.util.logging import InvokeAILogger
from invokeai.version.invokeai_version import __version__
from ..services.default_graphs import create_system_graphs
from ..services.latent_storage import DiskLatentsStorage, ForwardCacheLatentsStorage
from ..services.restoration_services import RestorationServices
from ..services.graph import GraphExecutionState, LibraryGraph
from ..services.image_file_storage import DiskImageFileStorage
from ..services.invocation_queue import MemoryInvocationQueue
@ -57,8 +57,9 @@ class ApiDependencies:
invoker: Invoker = None
@staticmethod
def initialize(config, event_handler_id: int, logger: Logger = logger):
logger.info(f"Internet connectivity is {config.internet_available}")
def initialize(config: InvokeAIAppConfig, event_handler_id: int, logger: Logger = logger):
logger.debug(f"InvokeAI version {__version__}")
logger.debug(f"Internet connectivity is {config.internet_available}")
events = FastAPIEventService(event_handler_id)
@ -73,7 +74,6 @@ class ApiDependencies:
)
urls = LocalUrlService()
metadata = CoreMetadataService()
image_record_storage = SqliteImageRecordStorage(db_location)
image_file_storage = DiskImageFileStorage(f"{output_folder}/images")
names = SimpleNameService()
@ -109,7 +109,6 @@ class ApiDependencies:
board_image_record_storage=board_image_record_storage,
image_record_storage=image_record_storage,
image_file_storage=image_file_storage,
metadata=metadata,
url=urls,
logger=logger,
names=names,
@ -118,7 +117,7 @@ class ApiDependencies:
)
services = InvocationServices(
model_manager=ModelManagerService(config,logger),
model_manager=ModelManagerService(config, logger),
events=events,
latents=latents,
images=images,
@ -130,7 +129,6 @@ class ApiDependencies:
),
graph_execution_manager=graph_execution_manager,
processor=DefaultInvocationProcessor(),
restoration=RestorationServices(config, logger),
configuration=config,
logger=logger,
)

36
invokeai/app/api/routers/app_info.py Normal file
View File

@ -0,0 +1,36 @@
from fastapi.routing import APIRouter
from pydantic import BaseModel, Field
from invokeai.backend.image_util.patchmatch import PatchMatch
from invokeai.version import __version__
app_router = APIRouter(prefix="/v1/app", tags=["app"])
class AppVersion(BaseModel):
"""App Version Response"""
version: str = Field(description="App version")
class AppConfig(BaseModel):
"""App Config Response"""
infill_methods: list[str] = Field(description="List of available infill methods")
@app_router.get(
"/version", operation_id="app_version", status_code=200, response_model=AppVersion
)
async def get_version() -> AppVersion:
return AppVersion(version=__version__)
@app_router.get(
"/config", operation_id="get_config", status_code=200, response_model=AppConfig
)
async def get_config() -> AppConfig:
infill_methods = ['tile']
if PatchMatch.patchmatch_available():
infill_methods.append('patchmatch')
return AppConfig(infill_methods=infill_methods)

63
invokeai/app/api/routers/images.py
View File

@ -1,25 +1,27 @@
import io
from typing import Optional
from fastapi import Body, HTTPException, Path, Query, Request, Response, UploadFile
from fastapi.routing import APIRouter
from fastapi import (Body, HTTPException, Path, Query, Request, Response,
UploadFile)
from fastapi.responses import FileResponse
from fastapi.routing import APIRouter
from PIL import Image
from invokeai.app.models.image import (
ImageCategory,
ResourceOrigin,
)
from invokeai.app.invocations.metadata import ImageMetadata
from invokeai.app.models.image import ImageCategory, ResourceOrigin
from invokeai.app.services.image_record_storage import OffsetPaginatedResults
from invokeai.app.services.models.image_record import (
ImageDTO,
ImageRecordChanges,
ImageUrlsDTO,
)
from invokeai.app.services.item_storage import PaginatedResults
from invokeai.app.services.models.image_record import (ImageDTO,
ImageRecordChanges,
ImageUrlsDTO)
from ..dependencies import ApiDependencies
images_router = APIRouter(prefix="/v1/images", tags=["images"])
# images are immutable; set a high max-age
IMAGE_MAX_AGE = 31536000
@images_router.post(
"/",
@ -103,23 +105,38 @@ async def update_image(
@images_router.get(
"/{image_name}/metadata",
operation_id="get_image_metadata",
"/{image_name}",
operation_id="get_image_dto",
response_model=ImageDTO,
)
async def get_image_metadata(
async def get_image_dto(
image_name: str = Path(description="The name of image to get"),
) -> ImageDTO:
"""Gets an image's metadata"""
"""Gets an image's DTO"""
try:
return ApiDependencies.invoker.services.images.get_dto(image_name)
except Exception as e:
raise HTTPException(status_code=404)
@images_router.get(
"/{image_name}/metadata",
operation_id="get_image_metadata",
response_model=ImageMetadata,
)
async def get_image_metadata(
image_name: str = Path(description="The name of image to get"),
) -> ImageMetadata:
"""Gets an image's metadata"""
try:
return ApiDependencies.invoker.services.images.get_metadata(image_name)
except Exception as e:
raise HTTPException(status_code=404)
@images_router.get(
"/{image_name}",
"/{image_name}/full",
operation_id="get_image_full",
response_class=Response,
responses={
@ -141,12 +158,14 @@ async def get_image_full(
if not ApiDependencies.invoker.services.images.validate_path(path):
raise HTTPException(status_code=404)
return FileResponse(
response = FileResponse(
path,
media_type="image/png",
filename=image_name,
content_disposition_type="inline",
)
response.headers["Cache-Control"] = f"max-age={IMAGE_MAX_AGE}"
return response
except Exception as e:
raise HTTPException(status_code=404)
@ -175,9 +194,11 @@ async def get_image_thumbnail(
if not ApiDependencies.invoker.services.images.validate_path(path):
raise HTTPException(status_code=404)
return FileResponse(
response = FileResponse(
path, media_type="image/webp", content_disposition_type="inline"
)
response.headers["Cache-Control"] = f"max-age={IMAGE_MAX_AGE}"
return response
except Exception as e:
raise HTTPException(status_code=404)
@ -208,10 +229,10 @@ async def get_image_urls(
@images_router.get(
"/",
operation_id="list_images_with_metadata",
operation_id="list_image_dtos",
response_model=OffsetPaginatedResults[ImageDTO],
)
async def list_images_with_metadata(
async def list_image_dtos(
image_origin: Optional[ResourceOrigin] = Query(
default=None, description="The origin of images to list"
),
@ -227,7 +248,7 @@ async def list_images_with_metadata(
offset: int = Query(default=0, description="The page offset"),
limit: int = Query(default=10, description="The number of images per page"),
) -> OffsetPaginatedResults[ImageDTO]:
"""Gets a list of images"""
"""Gets a list of image DTOs"""
image_dtos = ApiDependencies.invoker.services.images.get_many(
offset,

593
invokeai/app/api/routers/models.py
View File

@ -1,72 +1,34 @@
# Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654) and 2023 Kent Keirsey (https://github.com/hipsterusername)
# Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654), 2023 Kent Keirsey (https://github.com/hipsterusername), 2023 Lincoln D. Stein
from typing import Literal, Optional, Union
from fastapi import Query
from fastapi.routing import APIRouter, HTTPException
from pydantic import BaseModel, Field, parse_obj_as
from ..dependencies import ApiDependencies
import pathlib
from typing import Literal, List, Optional, Union
from fastapi import Body, Path, Query, Response
from fastapi.routing import APIRouter
from pydantic import BaseModel, parse_obj_as
from starlette.exceptions import HTTPException
from invokeai.backend import BaseModelType, ModelType
from invokeai.backend.model_management.models import OPENAPI_MODEL_CONFIGS, SchedulerPredictionType
MODEL_CONFIGS = Union[tuple(OPENAPI_MODEL_CONFIGS)]
from invokeai.backend.model_management.models import (
OPENAPI_MODEL_CONFIGS,
SchedulerPredictionType,
ModelNotFoundException,
)
from invokeai.backend.model_management import MergeInterpolationMethod
from ..dependencies import ApiDependencies
models_router = APIRouter(prefix="/v1/models", tags=["models"])
class VaeRepo(BaseModel):
repo_id: str = Field(description="The repo ID to use for this VAE")
path: Optional[str] = Field(description="The path to the VAE")
subfolder: Optional[str] = Field(description="The subfolder to use for this VAE")
class ModelInfo(BaseModel):
description: Optional[str] = Field(description="A description of the model")
model_name: str = Field(description="The name of the model")
model_type: str = Field(description="The type of the model")
class DiffusersModelInfo(ModelInfo):
format: Literal['folder'] = 'folder'
vae: Optional[VaeRepo] = Field(description="The VAE repo to use for this model")
repo_id: Optional[str] = Field(description="The repo ID to use for this model")
path: Optional[str] = Field(description="The path to the model")
class CkptModelInfo(ModelInfo):
format: Literal['ckpt'] = 'ckpt'
config: str = Field(description="The path to the model config")
weights: str = Field(description="The path to the model weights")
vae: str = Field(description="The path to the model VAE")
width: Optional[int] = Field(description="The width of the model")
height: Optional[int] = Field(description="The height of the model")
class SafetensorsModelInfo(CkptModelInfo):
format: Literal['safetensors'] = 'safetensors'
class CreateModelRequest(BaseModel):
name: str = Field(description="The name of the model")
info: Union[CkptModelInfo, DiffusersModelInfo] = Field(discriminator="format", description="The model info")
class CreateModelResponse(BaseModel):
name: str = Field(description="The name of the new model")
info: Union[CkptModelInfo, DiffusersModelInfo] = Field(discriminator="format", description="The model info")
status: str = Field(description="The status of the API response")
class ImportModelRequest(BaseModel):
name: str = Field(description="A model path, repo_id or URL to import")
prediction_type: Optional[Literal['epsilon','v_prediction','sample']] = Field(description='Prediction type for SDv2 checkpoint files')
class ConversionRequest(BaseModel):
name: str = Field(description="The name of the new model")
info: CkptModelInfo = Field(description="The converted model info")
save_location: str = Field(description="The path to save the converted model weights")
class ConvertedModelResponse(BaseModel):
name: str = Field(description="The name of the new model")
info: DiffusersModelInfo = Field(description="The converted model info")
UpdateModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
ImportModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
ConvertModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
MergeModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
ImportModelAttributes = Union[tuple(OPENAPI_MODEL_CONFIGS)]
class ModelsList(BaseModel):
models: list[MODEL_CONFIGS]
models: list[Union[tuple(OPENAPI_MODEL_CONFIGS)]]
@models_router.get(
"/",
@ -74,213 +36,362 @@ class ModelsList(BaseModel):
responses={200: {"model": ModelsList }},
)
async def list_models(
base_model: Optional[BaseModelType] = Query(
default=None, description="Base model"
),
model_type: Optional[ModelType] = Query(
default=None, description="The type of model to get"
),
base_model: Optional[BaseModelType] = Query(default=None, description="Base model"),
model_type: Optional[ModelType] = Query(default=None, description="The type of model to get"),
) -> ModelsList:
"""Gets a list of models"""
models_raw = ApiDependencies.invoker.services.model_manager.list_models(base_model, model_type)
models = parse_obj_as(ModelsList, { "models": models_raw })
return models
@models_router.post(
"/",
@models_router.patch(
"/{base_model}/{model_type}/{model_name}",
operation_id="update_model",
responses={200: {"status": "success"}},
responses={200: {"description" : "The model was updated successfully"},
400: {"description" : "Bad request"},
404: {"description" : "The model could not be found"},
409: {"description" : "There is already a model corresponding to the new name"},
},
status_code = 200,
response_model = UpdateModelResponse,
)
async def update_model(
model_request: CreateModelRequest
) -> CreateModelResponse:
""" Add Model """
model_request_info = model_request.info
info_dict = model_request_info.dict()
model_response = CreateModelResponse(name=model_request.name, info=model_request.info, status="success")
ApiDependencies.invoker.services.model_manager.add_model(
model_name=model_request.name,
model_attributes=info_dict,
clobber=True,
)
base_model: BaseModelType = Path(description="Base model"),
model_type: ModelType = Path(description="The type of model"),
model_name: str = Path(description="model name"),
info: Union[tuple(OPENAPI_MODEL_CONFIGS)] = Body(description="Model configuration"),
) -> UpdateModelResponse:
""" Update model contents with a new config. If the model name or base fields are changed, then the model is renamed. """
logger = ApiDependencies.invoker.services.logger
try:
# rename operation requested
if info.model_name != model_name or info.base_model != base_model:
result = ApiDependencies.invoker.services.model_manager.rename_model(
base_model = base_model,
model_type = model_type,
model_name = model_name,
new_name = info.model_name,
new_base = info.base_model,
)
logger.debug(f'renaming result = {result}')
logger.info(f'Successfully renamed {base_model}/{model_name}=>{info.base_model}/{info.model_name}')
model_name = info.model_name
base_model = info.base_model
ApiDependencies.invoker.services.model_manager.update_model(
model_name=model_name,
base_model=base_model,
model_type=model_type,
model_attributes=info.dict()
)
model_raw = ApiDependencies.invoker.services.model_manager.list_model(
model_name=model_name,
base_model=base_model,
model_type=model_type,
)
model_response = parse_obj_as(UpdateModelResponse, model_raw)
except ModelNotFoundException as e:
raise HTTPException(status_code=404, detail=str(e))
except ValueError as e:
logger.error(str(e))
raise HTTPException(status_code=409, detail=str(e))
except Exception as e:
logger.error(str(e))
raise HTTPException(status_code=400, detail=str(e))
return model_response
@models_router.post(
"/",
"/import",
operation_id="import_model",
responses={200: {"status": "success"}},
responses= {
201: {"description" : "The model imported successfully"},
404: {"description" : "The model could not be found"},
424: {"description" : "The model appeared to import successfully, but could not be found in the model manager"},
409: {"description" : "There is already a model corresponding to this path or repo_id"},
},
status_code=201,
response_model=ImportModelResponse
)
async def import_model(
model_request: ImportModelRequest
) -> None:
""" Add Model """
items_to_import = set([model_request.name])
location: str = Body(description="A model path, repo_id or URL to import"),
prediction_type: Optional[Literal['v_prediction','epsilon','sample']] = \
Body(description='Prediction type for SDv2 checkpoint files', default="v_prediction"),
) -> ImportModelResponse:
""" Add a model using its local path, repo_id, or remote URL. Model characteristics will be probed and configured automatically """
items_to_import = {location}
prediction_types = { x.value: x for x in SchedulerPredictionType }
logger = ApiDependencies.invoker.services.logger
installed_models = ApiDependencies.invoker.services.model_manager.heuristic_import(
items_to_import = items_to_import,
prediction_type_helper = lambda x: prediction_types.get(model_request.prediction_type)
)
if len(installed_models) > 0:
logger.info(f'Successfully imported {model_request.name}')
else:
logger.error(f'Model {model_request.name} not imported')
raise HTTPException(status_code=500, detail=f'Model {model_request.name} not imported')
try:
installed_models = ApiDependencies.invoker.services.model_manager.heuristic_import(
items_to_import = items_to_import,
prediction_type_helper = lambda x: prediction_types.get(prediction_type)
)
info = installed_models.get(location)
if not info:
logger.error("Import failed")
raise HTTPException(status_code=424)
logger.info(f'Successfully imported {location}, got {info}')
model_raw = ApiDependencies.invoker.services.model_manager.list_model(
model_name=info.name,
base_model=info.base_model,
model_type=info.model_type
)
return parse_obj_as(ImportModelResponse, model_raw)
except ModelNotFoundException as e:
logger.error(str(e))
raise HTTPException(status_code=404, detail=str(e))
except ValueError as e:
logger.error(str(e))
raise HTTPException(status_code=409, detail=str(e))
@models_router.post(
"/add",
operation_id="add_model",
responses= {
201: {"description" : "The model added successfully"},
404: {"description" : "The model could not be found"},
424: {"description" : "The model appeared to add successfully, but could not be found in the model manager"},
409: {"description" : "There is already a model corresponding to this path or repo_id"},
},
status_code=201,
response_model=ImportModelResponse
)
async def add_model(
info: Union[tuple(OPENAPI_MODEL_CONFIGS)] = Body(description="Model configuration"),
) -> ImportModelResponse:
""" Add a model using the configuration information appropriate for its type. Only local models can be added by path"""
logger = ApiDependencies.invoker.services.logger
try:
ApiDependencies.invoker.services.model_manager.add_model(
info.model_name,
info.base_model,
info.model_type,
model_attributes = info.dict()
)
logger.info(f'Successfully added {info.model_name}')
model_raw = ApiDependencies.invoker.services.model_manager.list_model(
model_name=info.model_name,
base_model=info.base_model,
model_type=info.model_type
)
return parse_obj_as(ImportModelResponse, model_raw)
except ModelNotFoundException as e:
logger.error(str(e))
raise HTTPException(status_code=404, detail=str(e))
except ValueError as e:
logger.error(str(e))
raise HTTPException(status_code=409, detail=str(e))
@models_router.delete(
"/{model_name}",
"/{base_model}/{model_type}/{model_name}",
operation_id="del_model",
responses={
204: {
"description": "Model deleted successfully"
},
404: {
"description": "Model not found"
}
204: { "description": "Model deleted successfully" },
404: { "description": "Model not found" }
},
status_code = 204,
response_model = None,
)
async def delete_model(model_name: str) -> None:
async def delete_model(
base_model: BaseModelType = Path(description="Base model"),
model_type: ModelType = Path(description="The type of model"),
model_name: str = Path(description="model name"),
) -> Response:
"""Delete Model"""
model_names = ApiDependencies.invoker.services.model_manager.model_names()
logger = ApiDependencies.invoker.services.logger
model_exists = model_name in model_names
# check if model exists
logger.info(f"Checking for model {model_name}...")
if model_exists:
logger.info(f"Deleting Model: {model_name}")
ApiDependencies.invoker.services.model_manager.del_model(model_name, delete_files=True)
logger.info(f"Model Deleted: {model_name}")
raise HTTPException(status_code=204, detail=f"Model '{model_name}' deleted successfully")
else:
logger.error("Model not found")
raise HTTPException(status_code=404, detail=f"Model '{model_name}' not found")
try:
ApiDependencies.invoker.services.model_manager.del_model(model_name,
base_model = base_model,
model_type = model_type
)
logger.info(f"Deleted model: {model_name}")
return Response(status_code=204)
except ModelNotFoundException as e:
logger.error(str(e))
raise HTTPException(status_code=404, detail=str(e))
@models_router.put(
"/convert/{base_model}/{model_type}/{model_name}",
operation_id="convert_model",
responses={
200: { "description": "Model converted successfully" },
400: {"description" : "Bad request" },
404: { "description": "Model not found" },
},
status_code = 200,
response_model = ConvertModelResponse,
)
async def convert_model(
base_model: BaseModelType = Path(description="Base model"),
model_type: ModelType = Path(description="The type of model"),
model_name: str = Path(description="model name"),
convert_dest_directory: Optional[str] = Query(default=None, description="Save the converted model to the designated directory"),
) -> ConvertModelResponse:
"""Convert a checkpoint model into a diffusers model, optionally saving to the indicated destination directory, or `models` if none."""
logger = ApiDependencies.invoker.services.logger
try:
logger.info(f"Converting model: {model_name}")
dest = pathlib.Path(convert_dest_directory) if convert_dest_directory else None
ApiDependencies.invoker.services.model_manager.convert_model(model_name,
base_model = base_model,
model_type = model_type,
convert_dest_directory = dest,
)
model_raw = ApiDependencies.invoker.services.model_manager.list_model(model_name,
base_model = base_model,
model_type = model_type)
response = parse_obj_as(ConvertModelResponse, model_raw)
except ModelNotFoundException as e:
raise HTTPException(status_code=404, detail=f"Model '{model_name}' not found: {str(e)}")
except ValueError as e:
raise HTTPException(status_code=400, detail=str(e))
return response
@models_router.get(
"/search",
operation_id="search_for_models",
responses={
200: { "description": "Directory searched successfully" },
404: { "description": "Invalid directory path" },
},
status_code = 200,
response_model = List[pathlib.Path]
)
async def search_for_models(
search_path: pathlib.Path = Query(description="Directory path to search for models")
)->List[pathlib.Path]:
if not search_path.is_dir():
raise HTTPException(status_code=404, detail=f"The search path '{search_path}' does not exist or is not directory")
return ApiDependencies.invoker.services.model_manager.search_for_models([search_path])
@models_router.get(
"/ckpt_confs",
operation_id="list_ckpt_configs",
responses={
200: { "description" : "paths retrieved successfully" },
},
status_code = 200,
response_model = List[pathlib.Path]
)
async def list_ckpt_configs(
)->List[pathlib.Path]:
"""Return a list of the legacy checkpoint configuration files stored in `ROOT/configs/stable-diffusion`, relative to ROOT."""
return ApiDependencies.invoker.services.model_manager.list_checkpoint_configs()
@models_router.get(
"/sync",
operation_id="sync_to_config",
responses={
201: { "description": "synchronization successful" },
},
status_code = 201,
response_model = None
)
async def sync_to_config(
)->None:
"""Call after making changes to models.yaml, autoimport directories or models directory to synchronize
in-memory data structures with disk data structures."""
return ApiDependencies.invoker.services.model_manager.sync_to_config()
@models_router.put(
"/merge/{base_model}",
operation_id="merge_models",
responses={
200: { "description": "Model converted successfully" },
400: { "description": "Incompatible models" },
404: { "description": "One or more models not found" },
},
status_code = 200,
response_model = MergeModelResponse,
)
async def merge_models(
base_model: BaseModelType = Path(description="Base model"),
model_names: List[str] = Body(description="model name", min_items=2, max_items=3),
merged_model_name: Optional[str] = Body(description="Name of destination model"),
alpha: Optional[float] = Body(description="Alpha weighting strength to apply to 2d and 3d models", default=0.5),
interp: Optional[MergeInterpolationMethod] = Body(description="Interpolation method"),
force: Optional[bool] = Body(description="Force merging of models created with different versions of diffusers", default=False),
merge_dest_directory: Optional[str] = Body(description="Save the merged model to the designated directory (with 'merged_model_name' appended)", default=None)
) -> MergeModelResponse:
"""Convert a checkpoint model into a diffusers model"""
logger = ApiDependencies.invoker.services.logger
try:
logger.info(f"Merging models: {model_names} into {merge_dest_directory or '<MODELS>'}/{merged_model_name}")
dest = pathlib.Path(merge_dest_directory) if merge_dest_directory else None
result = ApiDependencies.invoker.services.model_manager.merge_models(model_names,
base_model,
merged_model_name=merged_model_name or "+".join(model_names),
alpha=alpha,
interp=interp,
force=force,
merge_dest_directory = dest
)
model_raw = ApiDependencies.invoker.services.model_manager.list_model(result.name,
base_model = base_model,
model_type = ModelType.Main,
)
response = parse_obj_as(ConvertModelResponse, model_raw)
except ModelNotFoundException:
raise HTTPException(status_code=404, detail=f"One or more of the models '{model_names}' not found")
except ValueError as e:
raise HTTPException(status_code=400, detail=str(e))
return response
# @socketio.on("convertToDiffusers")
# def convert_to_diffusers(model_to_convert: dict):
# try:
# if model_info := self.generate.model_manager.model_info(
# model_name=model_to_convert["model_name"]
# ):
# if "weights" in model_info:
# ckpt_path = Path(model_info["weights"])
# original_config_file = Path(model_info["config"])
# model_name = model_to_convert["model_name"]
# model_description = model_info["description"]
# else:
# self.socketio.emit(
# "error", {"message": "Model is not a valid checkpoint file"}
# )
# else:
# self.socketio.emit(
# "error", {"message": "Could not retrieve model info."}
# )
# The rename operation is now supported by update_model and no longer needs to be
# a standalone route.
# @models_router.post(
# "/rename/{base_model}/{model_type}/{model_name}",
# operation_id="rename_model",
# responses= {
# 201: {"description" : "The model was renamed successfully"},
# 404: {"description" : "The model could not be found"},
# 409: {"description" : "There is already a model corresponding to the new name"},
# },
# status_code=201,
# response_model=ImportModelResponse
# )
# async def rename_model(
# base_model: BaseModelType = Path(description="Base model"),
# model_type: ModelType = Path(description="The type of model"),
# model_name: str = Path(description="current model name"),
# new_name: Optional[str] = Query(description="new model name", default=None),
# new_base: Optional[BaseModelType] = Query(description="new model base", default=None),
# ) -> ImportModelResponse:
# """ Rename a model"""
# logger = ApiDependencies.invoker.services.logger
# if not ckpt_path.is_absolute():
# ckpt_path = Path(Globals.root, ckpt_path)
# if original_config_file and not original_config_file.is_absolute():
# original_config_file = Path(Globals.root, original_config_file)
# diffusers_path = Path(
# ckpt_path.parent.absolute(), f"{model_name}_diffusers"
# )
# if model_to_convert["save_location"] == "root":
# diffusers_path = Path(
# global_converted_ckpts_dir(), f"{model_name}_diffusers"
# )
# if (
# model_to_convert["save_location"] == "custom"
# and model_to_convert["custom_location"] is not None
# ):
# diffusers_path = Path(
# model_to_convert["custom_location"], f"{model_name}_diffusers"
# )
# if diffusers_path.exists():
# shutil.rmtree(diffusers_path)
# self.generate.model_manager.convert_and_import(
# ckpt_path,
# diffusers_path,
# model_name=model_name,
# model_description=model_description,
# vae=None,
# original_config_file=original_config_file,
# commit_to_conf=opt.conf,
# )
# new_model_list = self.generate.model_manager.list_models()
# socketio.emit(
# "modelConverted",
# {
# "new_model_name": model_name,
# "model_list": new_model_list,
# "update": True,
# },
# )
# print(f">> Model Converted: {model_name}")
# except Exception as e:
# self.handle_exceptions(e)
# @socketio.on("mergeDiffusersModels")
# def merge_diffusers_models(model_merge_info: dict):
# try:
# models_to_merge = model_merge_info["models_to_merge"]
# model_ids_or_paths = [
# self.generate.model_manager.model_name_or_path(x)
# for x in models_to_merge
# ]
# merged_pipe = merge_diffusion_models(
# model_ids_or_paths,
# model_merge_info["alpha"],
# model_merge_info["interp"],
# model_merge_info["force"],
# )
# dump_path = global_models_dir() / "merged_models"
# if model_merge_info["model_merge_save_path"] is not None:
# dump_path = Path(model_merge_info["model_merge_save_path"])
# os.makedirs(dump_path, exist_ok=True)
# dump_path = dump_path / model_merge_info["merged_model_name"]
# merged_pipe.save_pretrained(dump_path, safe_serialization=1)
# merged_model_config = dict(
# model_name=model_merge_info["merged_model_name"],
# description=f'Merge of models {", ".join(models_to_merge)}',
# commit_to_conf=opt.conf,
# )
# if vae := self.generate.model_manager.config[models_to_merge[0]].get(
# "vae", None
# ):
# print(f">> Using configured VAE assigned to {models_to_merge[0]}")
# merged_model_config.update(vae=vae)
# self.generate.model_manager.import_diffuser_model(
# dump_path, **merged_model_config
# )
# new_model_list = self.generate.model_manager.list_models()
# socketio.emit(
# "modelsMerged",
# {
# "merged_models": models_to_merge,
# "merged_model_name": model_merge_info["merged_model_name"],
# "model_list": new_model_list,
# "update": True,
# },
# )
# print(f">> Models Merged: {models_to_merge}")
# print(f">> New Model Added: {model_merge_info['merged_model_name']}")
# except Exception as e:
# try:
# result = ApiDependencies.invoker.services.model_manager.rename_model(
# base_model = base_model,
# model_type = model_type,
# model_name = model_name,
# new_name = new_name,
# new_base = new_base,
# )
# logger.debug(result)
# logger.info(f'Successfully renamed {model_name}=>{new_name}')
# model_raw = ApiDependencies.invoker.services.model_manager.list_model(
# model_name=new_name or model_name,
# base_model=new_base or base_model,
# model_type=model_type
# )
# return parse_obj_as(ImportModelResponse, model_raw)
# except ModelNotFoundException as e:
# logger.error(str(e))
# raise HTTPException(status_code=404, detail=str(e))
# except ValueError as e:
# logger.error(str(e))
# raise HTTPException(status_code=409, detail=str(e))

26
invokeai/app/api_app.py
View File

@ -1,5 +1,6 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
# Copyright (c) 2022-2023 Kyle Schouviller (https://github.com/kyle0654) and the InvokeAI Team
import asyncio
import sys
from inspect import signature
import uvicorn
@ -20,13 +21,32 @@ from ..backend.util.logging import InvokeAILogger
app_config = InvokeAIAppConfig.get_config()
app_config.parse_args()
logger = InvokeAILogger.getLogger(config=app_config)
from invokeai.version.invokeai_version import __version__
# we call this early so that the message appears before
# other invokeai initialization messages
if app_config.version:
print(f'InvokeAI version {__version__}')
sys.exit(0)
import invokeai.frontend.web as web_dir
import mimetypes
from .api.dependencies import ApiDependencies
from .api.routers import sessions, models, images, boards, board_images
from .api.routers import sessions, models, images, boards, board_images, app_info
from .api.sockets import SocketIO
from .invocations.baseinvocation import BaseInvocation
import torch
import invokeai.backend.util.hotfixes
if torch.backends.mps.is_available():
import invokeai.backend.util.mps_fixes
# fix for windows mimetypes registry entries being borked
# see https://github.com/invoke-ai/InvokeAI/discussions/3684#discussioncomment-6391352
mimetypes.add_type('application/javascript', '.js')
mimetypes.add_type('text/css', '.css')
# Create the app
# TODO: create this all in a method so configuration/etc. can be passed in?
@ -82,6 +102,8 @@ app.include_router(boards.boards_router, prefix="/api")
app.include_router(board_images.board_images_router, prefix="/api")
app.include_router(app_info.app_router, prefix='/api')
# Build a custom OpenAPI to include all outputs
# TODO: can outputs be included on metadata of invocation schemas somehow?
def custom_openapi():

4
invokeai/app/cli/commands.py
View File

@ -47,7 +47,7 @@ def add_parsers(
commands: list[type],
command_field: str = "type",
exclude_fields: list[str] = ["id", "type"],
add_arguments: Callable[[argparse.ArgumentParser], None]|None = None
add_arguments: Union[Callable[[argparse.ArgumentParser], None],None] = None
):
"""Adds parsers for each command to the subparsers"""
@ -72,7 +72,7 @@ def add_parsers(
def add_graph_parsers(
subparsers,
graphs: list[LibraryGraph],
add_arguments: Callable[[argparse.ArgumentParser], None]|None = None
add_arguments: Union[Callable[[argparse.ArgumentParser], None], None] = None
):
for graph in graphs:
command_parser = subparsers.add_parser(graph.name, help=graph.description)

22
invokeai/app/cli_app.py
View File

@ -1,12 +1,11 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
import argparse
import os
import re
import shlex
import sys
import time
from typing import Union, get_type_hints
from typing import Union, get_type_hints, Optional
from pydantic import BaseModel, ValidationError
from pydantic.fields import Field
@ -17,6 +16,12 @@ from invokeai.backend.util.logging import InvokeAILogger
config = InvokeAIAppConfig.get_config()
config.parse_args()
logger = InvokeAILogger().getLogger(config=config)
from invokeai.version.invokeai_version import __version__
# we call this early so that the message appears before other invokeai initialization messages
if config.version:
print(f'InvokeAI version {__version__}')
sys.exit(0)
from invokeai.app.services.board_image_record_storage import (
SqliteBoardImageRecordStorage,
@ -29,7 +34,6 @@ from invokeai.app.services.board_record_storage import SqliteBoardRecordStorage
from invokeai.app.services.boards import BoardService, BoardServiceDependencies
from invokeai.app.services.image_record_storage import SqliteImageRecordStorage
from invokeai.app.services.images import ImageService, ImageServiceDependencies
from invokeai.app.services.metadata import CoreMetadataService
from invokeai.app.services.resource_name import SimpleNameService
from invokeai.app.services.urls import LocalUrlService
from .services.default_graphs import (default_text_to_image_graph_id,
@ -50,9 +54,13 @@ from .services.invocation_services import InvocationServices
from .services.invoker import Invoker
from .services.model_manager_service import ModelManagerService
from .services.processor import DefaultInvocationProcessor
from .services.restoration_services import RestorationServices
from .services.sqlite import SqliteItemStorage
import torch
import invokeai.backend.util.hotfixes
if torch.backends.mps.is_available():
import invokeai.backend.util.mps_fixes
class CliCommand(BaseModel):
command: Union[BaseCommand.get_commands() + BaseInvocation.get_invocations()] = Field(discriminator="type") # type: ignore
@ -205,6 +213,7 @@ def invoke_all(context: CliContext):
raise SessionError()
def invoke_cli():
logger.info(f'InvokeAI version {__version__}')
# get the optional list of invocations to execute on the command line
parser = config.get_parser()
parser.add_argument('commands',nargs='*')
@ -234,7 +243,6 @@ def invoke_cli():
)
urls = LocalUrlService()
metadata = CoreMetadataService()
image_record_storage = SqliteImageRecordStorage(db_location)
image_file_storage = DiskImageFileStorage(f"{output_folder}/images")
names = SimpleNameService()
@ -267,7 +275,6 @@ def invoke_cli():
board_image_record_storage=board_image_record_storage,
image_record_storage=image_record_storage,
image_file_storage=image_file_storage,
metadata=metadata,
url=urls,
logger=logger,
names=names,
@ -288,7 +295,6 @@ def invoke_cli():
),
graph_execution_manager=graph_execution_manager,
processor=DefaultInvocationProcessor(),
restoration=RestorationServices(config,logger=logger),
logger=logger,
configuration=config,
)
@ -348,7 +354,7 @@ def invoke_cli():
# Parse invocation
command: CliCommand = None # type:ignore
system_graph: LibraryGraph|None = None
system_graph: Optional[LibraryGraph] = None
if args['type'] in system_graph_names:
system_graph = next(filter(lambda g: g.name == args['type'], system_graphs))
invocation = GraphInvocation(graph=system_graph.graph, id=str(current_id))

25
invokeai/app/invocations/baseinvocation.py
View File

@ -4,9 +4,10 @@ from __future__ import annotations
from abc import ABC, abstractmethod
from inspect import signature
from typing import get_args, get_type_hints, Dict, List, Literal, TypedDict, TYPE_CHECKING
from typing import (TYPE_CHECKING, Dict, List, Literal, TypedDict, get_args,
get_type_hints)
from pydantic import BaseModel, Field
from pydantic import BaseConfig, BaseModel, Field
if TYPE_CHECKING:
from ..services.invocation_services import InvocationServices
@ -65,8 +66,13 @@ class BaseInvocation(ABC, BaseModel):
@classmethod
def get_invocations_map(cls):
# Get the type strings out of the literals and into a dictionary
return dict(map(lambda t: (get_args(get_type_hints(t)['type'])[0], t),BaseInvocation.get_all_subclasses()))
return dict(
map(
lambda t: (get_args(get_type_hints(t)["type"])[0], t),
BaseInvocation.get_all_subclasses(),
)
)
@classmethod
def get_output_type(cls):
return signature(cls.invoke).return_annotation
@ -75,11 +81,11 @@ class BaseInvocation(ABC, BaseModel):
def invoke(self, context: InvocationContext) -> BaseInvocationOutput:
"""Invoke with provided context and return outputs."""
pass
#fmt: off
# fmt: off
id: str = Field(description="The id of this node. Must be unique among all nodes.")
is_intermediate: bool = Field(default=False, description="Whether or not this node is an intermediate node.")
#fmt: on
# fmt: on
# TODO: figure out a better way to provide these hints
@ -98,16 +104,19 @@ class UIConfig(TypedDict, total=False):
"model",
"control",
"image_collection",
"vae_model",
"lora_model",
],
]
tags: List[str]
title: str
class CustomisedSchemaExtra(TypedDict):
ui: UIConfig
class InvocationConfig(BaseModel.Config):
class InvocationConfig(BaseConfig):
"""Customizes pydantic's BaseModel.Config class for use by Invocations.
Provide `schema_extra` a `ui` dict to add hints for generated UIs.

182
invokeai/app/invocations/compel.py
View File

@ -1,27 +1,25 @@
from typing import Literal, Optional, Union
from typing import Literal, Optional, Union, List
from pydantic import BaseModel, Field
from contextlib import ExitStack
import re
from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext, InvocationConfig
from .model import ClipField
from ...backend.util.devices import torch_dtype
from ...backend.stable_diffusion.diffusion import InvokeAIDiffuserComponent
from ...backend.model_management import BaseModelType, ModelType, SubModelType
from ...backend.model_management.lora import ModelPatcher
import torch
from compel import Compel
from compel.prompt_parser import (
Blend,
CrossAttentionControlSubstitute,
FlattenedPrompt,
Fragment, Conjunction,
)
from compel.prompt_parser import (Blend, Conjunction,
CrossAttentionControlSubstitute,
FlattenedPrompt, Fragment)
from ...backend.util.devices import torch_dtype
from ...backend.model_management import ModelType
from ...backend.model_management.models import ModelNotFoundException
from ...backend.model_management.lora import ModelPatcher
from ...backend.stable_diffusion.diffusion import InvokeAIDiffuserComponent
from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
InvocationConfig, InvocationContext)
from .model import ClipField
class ConditioningField(BaseModel):
conditioning_name: Optional[str] = Field(default=None, description="The name of conditioning data")
conditioning_name: Optional[str] = Field(
default=None, description="The name of conditioning data")
class Config:
schema_extra = {"required": ["conditioning_name"]}
@ -51,83 +49,108 @@ class CompelInvocation(BaseInvocation):
"title": "Prompt (Compel)",
"tags": ["prompt", "compel"],
"type_hints": {
"model": "model"
"model": "model"
}
},
}
@torch.no_grad()
def invoke(self, context: InvocationContext) -> CompelOutput:
tokenizer_info = context.services.model_manager.get_model(
**self.clip.tokenizer.dict(),
)
text_encoder_info = context.services.model_manager.get_model(
**self.clip.text_encoder.dict(),
)
with tokenizer_info as orig_tokenizer,\
text_encoder_info as text_encoder:
loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.clip.loras]
def _lora_loader():
for lora in self.clip.loras:
lora_info = context.services.model_manager.get_model(
**lora.dict(exclude={"weight"}))
yield (lora_info.context.model, lora.weight)
del lora_info
return
ti_list = []
for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", self.prompt):
name = trigger[1:-1]
try:
ti_list.append(
context.services.model_manager.get_model(
model_name=name,
base_model=self.clip.text_encoder.base_model,
model_type=ModelType.TextualInversion,
).context.model
)
except Exception:
#print(e)
#import traceback
#print(traceback.format_exc())
print(f"Warn: trigger: \"{trigger}\" not found")
#loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.clip.loras]
with ModelPatcher.apply_lora_text_encoder(text_encoder, loras),\
ModelPatcher.apply_ti(orig_tokenizer, text_encoder, ti_list) as (tokenizer, ti_manager):
compel = Compel(
tokenizer=tokenizer,
text_encoder=text_encoder,
textual_inversion_manager=ti_manager,
dtype_for_device_getter=torch_dtype,
truncate_long_prompts=True, # TODO:
ti_list = []
for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", self.prompt):
name = trigger[1:-1]
try:
ti_list.append(
context.services.model_manager.get_model(
model_name=name,
base_model=self.clip.text_encoder.base_model,
model_type=ModelType.TextualInversion,
).context.model
)
conjunction = Compel.parse_prompt_string(self.prompt)
prompt: Union[FlattenedPrompt, Blend] = conjunction.prompts[0]
except ModelNotFoundException:
# print(e)
#import traceback
#print(traceback.format_exc())
print(f"Warn: trigger: \"{trigger}\" not found")
if context.services.configuration.log_tokenization:
log_tokenization_for_prompt_object(prompt, tokenizer)
with ModelPatcher.apply_lora_text_encoder(text_encoder_info.context.model, _lora_loader()),\
ModelPatcher.apply_ti(tokenizer_info.context.model, text_encoder_info.context.model, ti_list) as (tokenizer, ti_manager),\
ModelPatcher.apply_clip_skip(text_encoder_info.context.model, self.clip.skipped_layers),\
text_encoder_info as text_encoder:
c, options = compel.build_conditioning_tensor_for_prompt_object(prompt)
# TODO: long prompt support
#if not self.truncate_long_prompts:
# [c, uc] = compel.pad_conditioning_tensors_to_same_length([c, uc])
ec = InvokeAIDiffuserComponent.ExtraConditioningInfo(
tokens_count_including_eos_bos=get_max_token_count(tokenizer, conjunction),
cross_attention_control_args=options.get("cross_attention_control", None),
)
conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
# TODO: hacky but works ;D maybe rename latents somehow?
context.services.latents.save(conditioning_name, (c, ec))
return CompelOutput(
conditioning=ConditioningField(
conditioning_name=conditioning_name,
),
compel = Compel(
tokenizer=tokenizer,
text_encoder=text_encoder,
textual_inversion_manager=ti_manager,
dtype_for_device_getter=torch_dtype,
truncate_long_prompts=False,
)
conjunction = Compel.parse_prompt_string(self.prompt)
prompt: Union[FlattenedPrompt, Blend] = conjunction.prompts[0]
if context.services.configuration.log_tokenization:
log_tokenization_for_prompt_object(prompt, tokenizer)
c, options = compel.build_conditioning_tensor_for_prompt_object(
prompt)
ec = InvokeAIDiffuserComponent.ExtraConditioningInfo(
tokens_count_including_eos_bos=get_max_token_count(
tokenizer, conjunction),
cross_attention_control_args=options.get(
"cross_attention_control", None),)
conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
# TODO: hacky but works ;D maybe rename latents somehow?
context.services.latents.save(conditioning_name, (c, ec))
return CompelOutput(
conditioning=ConditioningField(
conditioning_name=conditioning_name,
),
)
class ClipSkipInvocationOutput(BaseInvocationOutput):
"""Clip skip node output"""
type: Literal["clip_skip_output"] = "clip_skip_output"
clip: ClipField = Field(None, description="Clip with skipped layers")
class ClipSkipInvocation(BaseInvocation):
"""Skip layers in clip text_encoder model."""
type: Literal["clip_skip"] = "clip_skip"
clip: ClipField = Field(None, description="Clip to use")
skipped_layers: int = Field(0, description="Number of layers to skip in text_encoder")
def invoke(self, context: InvocationContext) -> ClipSkipInvocationOutput:
self.clip.skipped_layers += self.skipped_layers
return ClipSkipInvocationOutput(
clip=self.clip,
)
def get_max_token_count(
tokenizer, prompt: Union[FlattenedPrompt, Blend, Conjunction], truncate_if_too_long=False
) -> int:
tokenizer, prompt: Union[FlattenedPrompt, Blend, Conjunction],
truncate_if_too_long=False) -> int:
if type(prompt) is Blend:
blend: Blend = prompt
return max(
@ -146,13 +169,13 @@ def get_max_token_count(
)
else:
return len(
get_tokens_for_prompt_object(tokenizer, prompt, truncate_if_too_long)
)
get_tokens_for_prompt_object(
tokenizer, prompt, truncate_if_too_long))
def get_tokens_for_prompt_object(
tokenizer, parsed_prompt: FlattenedPrompt, truncate_if_too_long=True
) -> [str]:
) -> List[str]:
if type(parsed_prompt) is Blend:
raise ValueError(
"Blend is not supported here - you need to get tokens for each of its .children"
@ -181,7 +204,7 @@ def log_tokenization_for_conjunction(
):
display_label_prefix = display_label_prefix or ""
for i, p in enumerate(c.prompts):
if len(c.prompts)>1:
if len(c.prompts) > 1:
this_display_label_prefix = f"{display_label_prefix}(conjunction part {i + 1}, weight={c.weights[i]})"
else:
this_display_label_prefix = display_label_prefix
@ -236,7 +259,8 @@ def log_tokenization_for_prompt_object(
)
def log_tokenization_for_text(text, tokenizer, display_label=None, truncate_if_too_long=False):
def log_tokenization_for_text(
text, tokenizer, display_label=None, truncate_if_too_long=False):
"""shows how the prompt is tokenized
# usually tokens have '</w>' to indicate end-of-word,
# but for readability it has been replaced with ' '

34
invokeai/app/invocations/controlnet_image_processors.py
View File

@ -6,9 +6,10 @@ from builtins import float, bool
import cv2
import numpy as np
from typing import Literal, Optional, Union, List, Dict
from PIL import Image, ImageFilter, ImageOps
from PIL import Image
from pydantic import BaseModel, Field, validator
from ...backend.model_management import BaseModelType, ModelType
from ..models.image import ImageField, ImageCategory, ResourceOrigin
from .baseinvocation import (
BaseInvocation,
@ -105,9 +106,15 @@ CONTROLNET_MODE_VALUES = Literal[tuple(["balanced", "more_prompt", "more_control
# CONTROLNET_RESIZE_VALUES = Literal[tuple(["just_resize", "crop_resize", "fill_resize"])]
class ControlNetModelField(BaseModel):
"""ControlNet model field"""
model_name: str = Field(description="Name of the ControlNet model")
base_model: BaseModelType = Field(description="Base model")
class ControlField(BaseModel):
image: ImageField = Field(default=None, description="The control image")
control_model: Optional[str] = Field(default=None, description="The ControlNet model to use")
control_model: Optional[ControlNetModelField] = Field(default=None, description="The ControlNet model to use")
# control_weight: Optional[float] = Field(default=1, description="weight given to controlnet")
control_weight: Union[float, List[float]] = Field(default=1, description="The weight given to the ControlNet")
begin_step_percent: float = Field(default=0, ge=0, le=1,
@ -118,15 +125,15 @@ class ControlField(BaseModel):
# resize_mode: CONTROLNET_RESIZE_VALUES = Field(default="just_resize", description="The resize mode to use")
@validator("control_weight")
def abs_le_one(cls, v):
"""validate that all abs(values) are <=1"""
def validate_control_weight(cls, v):
"""Validate that all control weights in the valid range"""
if isinstance(v, list):
for i in v:
if abs(i) > 1:
raise ValueError('all abs(control_weight) must be <= 1')
if i < -1 or i > 2:
raise ValueError('Control weights must be within -1 to 2 range')
else:
if abs(v) > 1:
raise ValueError('abs(control_weight) must be <= 1')
if v < -1 or v > 2:
raise ValueError('Control weights must be within -1 to 2 range')
return v
class Config:
schema_extra = {
@ -134,6 +141,7 @@ class ControlField(BaseModel):
"ui": {
"type_hints": {
"control_weight": "float",
"control_model": "controlnet_model",
# "control_weight": "number",
}
}
@ -154,10 +162,10 @@ class ControlNetInvocation(BaseInvocation):
type: Literal["controlnet"] = "controlnet"
# Inputs
image: ImageField = Field(default=None, description="The control image")
control_model: CONTROLNET_NAME_VALUES = Field(default="lllyasviel/sd-controlnet-canny",
control_model: ControlNetModelField = Field(default="lllyasviel/sd-controlnet-canny",
description="control model used")
control_weight: Union[float, List[float]] = Field(default=1.0, description="The weight given to the ControlNet")
begin_step_percent: float = Field(default=0, ge=0, le=1,
begin_step_percent: float = Field(default=0, ge=-1, le=2,
description="When the ControlNet is first applied (% of total steps)")
end_step_percent: float = Field(default=1, ge=0, le=1,
description="When the ControlNet is last applied (% of total steps)")
@ -422,9 +430,9 @@ class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation, PILInvoca
# Inputs
detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
h: Union[int, None] = Field(default=512, ge=0, description="Content shuffle `h` parameter")
w: Union[int, None] = Field(default=512, ge=0, description="Content shuffle `w` parameter")
f: Union[int, None] = Field(default=256, ge=0, description="Content shuffle `f` parameter")
h: Optional[int] = Field(default=512, ge=0, description="Content shuffle `h` parameter")
w: Optional[int] = Field(default=512, ge=0, description="Content shuffle `w` parameter")
f: Optional[int] = Field(default=256, ge=0, description="Content shuffle `f` parameter")
# fmt: on
def run_processor(self, image):

100
invokeai/app/invocations/generate.py
View File

@ -1,11 +1,10 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
from functools import partial
from typing import Literal, Optional, Union, get_args
from typing import Literal, Optional, get_args
import torch
from diffusers import ControlNetModel
from pydantic import BaseModel, Field
from pydantic import Field
from invokeai.app.models.image import (ColorField, ImageCategory, ImageField,
ResourceOrigin)
@ -18,7 +17,6 @@ from ..util.step_callback import stable_diffusion_step_callback
from .baseinvocation import BaseInvocation, InvocationConfig, InvocationContext
from .image import ImageOutput
import re
from ...backend.model_management.lora import ModelPatcher
from ...backend.stable_diffusion.diffusers_pipeline import StableDiffusionGeneratorPipeline
from .model import UNetField, VaeField
@ -76,7 +74,7 @@ class InpaintInvocation(BaseInvocation):
vae: VaeField = Field(default=None, description="Vae model")
# Inputs
image: Union[ImageField, None] = Field(description="The input image")
image: Optional[ImageField] = Field(description="The input image")
strength: float = Field(
default=0.75, gt=0, le=1, description="The strength of the original image"
)
@ -86,7 +84,7 @@ class InpaintInvocation(BaseInvocation):
)
# Inputs
mask: Union[ImageField, None] = Field(description="The mask")
mask: Optional[ImageField] = Field(description="The mask")
seam_size: int = Field(default=96, ge=1, description="The seam inpaint size (px)")
seam_blur: int = Field(
default=16, ge=0, description="The seam inpaint blur radius (px)"
@ -156,40 +154,42 @@ class InpaintInvocation(BaseInvocation):
@contextmanager
def load_model_old_way(self, context, scheduler):
def _lora_loader():
for lora in self.unet.loras:
lora_info = context.services.model_manager.get_model(
**lora.dict(exclude={"weight"}))
yield (lora_info.context.model, lora.weight)
del lora_info
return
unet_info = context.services.model_manager.get_model(**self.unet.unet.dict())
vae_info = context.services.model_manager.get_model(**self.vae.vae.dict())
#unet = unet_info.context.model
#vae = vae_info.context.model
with vae_info as vae,\
ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
unet_info as unet:
with ExitStack() as stack:
loras = [(stack.enter_context(context.services.model_manager.get_model(**lora.dict(exclude={"weight"}))), lora.weight) for lora in self.unet.loras]
device = context.services.model_manager.mgr.cache.execution_device
dtype = context.services.model_manager.mgr.cache.precision
with vae_info as vae,\
unet_info as unet,\
ModelPatcher.apply_lora_unet(unet, loras):
pipeline = StableDiffusionGeneratorPipeline(
vae=vae,
text_encoder=None,
tokenizer=None,
unet=unet,
scheduler=scheduler,
safety_checker=None,
feature_extractor=None,
requires_safety_checker=False,
precision="float16" if dtype == torch.float16 else "float32",
execution_device=device,
)
device = context.services.model_manager.mgr.cache.execution_device
dtype = context.services.model_manager.mgr.cache.precision
pipeline = StableDiffusionGeneratorPipeline(
vae=vae,
text_encoder=None,
tokenizer=None,
unet=unet,
scheduler=scheduler,
safety_checker=None,
feature_extractor=None,
requires_safety_checker=False,
precision="float16" if dtype == torch.float16 else "float32",
execution_device=device,
)
yield OldModelInfo(
name=self.unet.unet.model_name,
hash="<NO-HASH>",
model=pipeline,
)
yield OldModelInfo(
name=self.unet.unet.model_name,
hash="<NO-HASH>",
model=pipeline,
)
def invoke(self, context: InvocationContext) -> ImageOutput:
image = (
@ -228,21 +228,21 @@ class InpaintInvocation(BaseInvocation):
), # Shorthand for passing all of the parameters above manually
)
# Outputs is an infinite iterator that will return a new InvokeAIGeneratorOutput object
# each time it is called. We only need the first one.
generator_output = next(outputs)
# Outputs is an infinite iterator that will return a new InvokeAIGeneratorOutput object
# each time it is called. We only need the first one.
generator_output = next(outputs)
image_dto = context.services.images.create(
image=generator_output.image,
image_origin=ResourceOrigin.INTERNAL,
image_category=ImageCategory.GENERAL,
session_id=context.graph_execution_state_id,
node_id=self.id,
is_intermediate=self.is_intermediate,
)
image_dto = context.services.images.create(
image=generator_output.image,
image_origin=ResourceOrigin.INTERNAL,
image_category=ImageCategory.GENERAL,
session_id=context.graph_execution_state_id,
node_id=self.id,
is_intermediate=self.is_intermediate,
)
return ImageOutput(
image=ImageField(image_name=image_dto.image_name),
width=image_dto.width,
height=image_dto.height,
)
return ImageOutput(
image=ImageField(image_name=image_dto.image_name),
width=image_dto.width,
height=image_dto.height,
)

38
invokeai/app/invocations/image.py
View File

@ -1,11 +1,11 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
import io
from typing import Literal, Optional, Union
from typing import Literal, Optional
import numpy
from PIL import Image, ImageFilter, ImageOps, ImageChops
from pydantic import BaseModel, Field
from typing import Union
from ..models.image import ImageCategory, ImageField, ResourceOrigin
from .baseinvocation import (
@ -67,7 +67,7 @@ class LoadImageInvocation(BaseInvocation):
type: Literal["load_image"] = "load_image"
# Inputs
image: Union[ImageField, None] = Field(
image: Optional[ImageField] = Field(
default=None, description="The image to load"
)
# fmt: on
@ -87,7 +87,7 @@ class ShowImageInvocation(BaseInvocation):
type: Literal["show_image"] = "show_image"
# Inputs
image: Union[ImageField, None] = Field(
image: Optional[ImageField] = Field(
default=None, description="The image to show"
)
@ -112,7 +112,7 @@ class ImageCropInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_crop"] = "img_crop"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to crop")
image: Optional[ImageField] = Field(default=None, description="The image to crop")
x: int = Field(default=0, description="The left x coordinate of the crop rectangle")
y: int = Field(default=0, description="The top y coordinate of the crop rectangle")
width: int = Field(default=512, gt=0, description="The width of the crop rectangle")
@ -150,8 +150,8 @@ class ImagePasteInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_paste"] = "img_paste"
# Inputs
base_image: Union[ImageField, None] = Field(default=None, description="The base image")
image: Union[ImageField, None] = Field(default=None, description="The image to paste")
base_image: Optional[ImageField] = Field(default=None, description="The base image")
image: Optional[ImageField] = Field(default=None, description="The image to paste")
mask: Optional[ImageField] = Field(default=None, description="The mask to use when pasting")
x: int = Field(default=0, description="The left x coordinate at which to paste the image")
y: int = Field(default=0, description="The top y coordinate at which to paste the image")
@ -203,7 +203,7 @@ class MaskFromAlphaInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["tomask"] = "tomask"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to create the mask from")
image: Optional[ImageField] = Field(default=None, description="The image to create the mask from")
invert: bool = Field(default=False, description="Whether or not to invert the mask")
# fmt: on
@ -237,8 +237,8 @@ class ImageMultiplyInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_mul"] = "img_mul"
# Inputs
image1: Union[ImageField, None] = Field(default=None, description="The first image to multiply")
image2: Union[ImageField, None] = Field(default=None, description="The second image to multiply")
image1: Optional[ImageField] = Field(default=None, description="The first image to multiply")
image2: Optional[ImageField] = Field(default=None, description="The second image to multiply")
# fmt: on
def invoke(self, context: InvocationContext) -> ImageOutput:
@ -273,7 +273,7 @@ class ImageChannelInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_chan"] = "img_chan"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to get the channel from")
image: Optional[ImageField] = Field(default=None, description="The image to get the channel from")
channel: IMAGE_CHANNELS = Field(default="A", description="The channel to get")
# fmt: on
@ -308,7 +308,7 @@ class ImageConvertInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_conv"] = "img_conv"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to convert")
image: Optional[ImageField] = Field(default=None, description="The image to convert")
mode: IMAGE_MODES = Field(default="L", description="The mode to convert to")
# fmt: on
@ -340,7 +340,7 @@ class ImageBlurInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_blur"] = "img_blur"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to blur")
image: Optional[ImageField] = Field(default=None, description="The image to blur")
radius: float = Field(default=8.0, ge=0, description="The blur radius")
blur_type: Literal["gaussian", "box"] = Field(default="gaussian", description="The type of blur")
# fmt: on
@ -398,9 +398,9 @@ class ImageResizeInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_resize"] = "img_resize"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to resize")
width: int = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
height: int = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
image: Optional[ImageField] = Field(default=None, description="The image to resize")
width: Union[int, None] = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
height: Union[int, None] = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
resample_mode: PIL_RESAMPLING_MODES = Field(default="bicubic", description="The resampling mode")
# fmt: on
@ -437,7 +437,7 @@ class ImageScaleInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_scale"] = "img_scale"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to scale")
image: Optional[ImageField] = Field(default=None, description="The image to scale")
scale_factor: float = Field(gt=0, description="The factor by which to scale the image")
resample_mode: PIL_RESAMPLING_MODES = Field(default="bicubic", description="The resampling mode")
# fmt: on
@ -477,7 +477,7 @@ class ImageLerpInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_lerp"] = "img_lerp"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to lerp")
image: Optional[ImageField] = Field(default=None, description="The image to lerp")
min: int = Field(default=0, ge=0, le=255, description="The minimum output value")
max: int = Field(default=255, ge=0, le=255, description="The maximum output value")
# fmt: on
@ -513,7 +513,7 @@ class ImageInverseLerpInvocation(BaseInvocation, PILInvocationConfig):
type: Literal["img_ilerp"] = "img_ilerp"
# Inputs
image: Union[ImageField, None] = Field(default=None, description="The image to lerp")
image: Optional[ImageField] = Field(default=None, description="The image to lerp")
min: int = Field(default=0, ge=0, le=255, description="The minimum input value")
max: int = Field(default=255, ge=0, le=255, description="The maximum input value")
# fmt: on

10
invokeai/app/invocations/infill.py
View File

@ -1,6 +1,6 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654) and the InvokeAI Team
from typing import Literal, Union, get_args
from typing import Literal, Optional, get_args
import numpy as np
import math
@ -68,7 +68,7 @@ def get_tile_images(image: np.ndarray, width=8, height=8):
def tile_fill_missing(
im: Image.Image, tile_size: int = 16, seed: Union[int, None] = None
im: Image.Image, tile_size: int = 16, seed: Optional[int] = None
) -> Image.Image:
# Only fill if there's an alpha layer
if im.mode != "RGBA":
@ -125,7 +125,7 @@ class InfillColorInvocation(BaseInvocation):
"""Infills transparent areas of an image with a solid color"""
type: Literal["infill_rgba"] = "infill_rgba"
image: Union[ImageField, None] = Field(
image: Optional[ImageField] = Field(
default=None, description="The image to infill"
)
color: ColorField = Field(
@ -162,7 +162,7 @@ class InfillTileInvocation(BaseInvocation):
type: Literal["infill_tile"] = "infill_tile"
image: Union[ImageField, None] = Field(
image: Optional[ImageField] = Field(
default=None, description="The image to infill"
)
tile_size: int = Field(default=32, ge=1, description="The tile size (px)")
@ -202,7 +202,7 @@ class InfillPatchMatchInvocation(BaseInvocation):
type: Literal["infill_patchmatch"] = "infill_patchmatch"
image: Union[ImageField, None] = Field(
image: Optional[ImageField] = Field(
default=None, description="The image to infill"
)

313
invokeai/app/invocations/latent.py
View File

@ -4,18 +4,17 @@ from contextlib import ExitStack
from typing import List, Literal, Optional, Union
import einops
from pydantic import BaseModel, Field, validator
import torch
from diffusers import ControlNetModel, DPMSolverMultistepScheduler
from diffusers import ControlNetModel
from diffusers.image_processor import VaeImageProcessor
from diffusers.schedulers import SchedulerMixin as Scheduler
from pydantic import BaseModel, Field, validator
from invokeai.app.util.misc import SEED_MAX, get_random_seed
from invokeai.app.invocations.metadata import CoreMetadata
from invokeai.app.util.step_callback import stable_diffusion_step_callback
from invokeai.backend.model_management.models.base import ModelType
from ..models.image import ImageCategory, ImageField, ResourceOrigin
from ...backend.image_util.seamless import configure_model_padding
from ...backend.model_management.lora import ModelPatcher
from ...backend.stable_diffusion import PipelineIntermediateState
from ...backend.stable_diffusion.diffusers_pipeline import (
ConditioningData, ControlNetData, StableDiffusionGeneratorPipeline,
@ -24,7 +23,7 @@ from ...backend.stable_diffusion.diffusion.shared_invokeai_diffusion import \
PostprocessingSettings
from ...backend.stable_diffusion.schedulers import SCHEDULER_MAP
from ...backend.util.devices import torch_dtype
from ...backend.model_management.lora import ModelPatcher
from ..models.image import ImageCategory, ImageField, ResourceOrigin
from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
InvocationConfig, InvocationContext)
from .compel import ConditioningField
@ -32,14 +31,17 @@ from .controlnet_image_processors import ControlField
from .image import ImageOutput
from .model import ModelInfo, UNetField, VaeField
class LatentsField(BaseModel):
"""A latents field used for passing latents between invocations"""
latents_name: Optional[str] = Field(default=None, description="The name of the latents")
latents_name: Optional[str] = Field(
default=None, description="The name of the latents")
class Config:
schema_extra = {"required": ["latents_name"]}
class LatentsOutput(BaseInvocationOutput):
"""Base class for invocations that output latents"""
#fmt: off
@ -53,11 +55,11 @@ class LatentsOutput(BaseInvocationOutput):
def build_latents_output(latents_name: str, latents: torch.Tensor):
return LatentsOutput(
latents=LatentsField(latents_name=latents_name),
width=latents.size()[3] * 8,
height=latents.size()[2] * 8,
)
return LatentsOutput(
latents=LatentsField(latents_name=latents_name),
width=latents.size()[3] * 8,
height=latents.size()[2] * 8,
)
SAMPLER_NAME_VALUES = Literal[
@ -70,16 +72,24 @@ def get_scheduler(
scheduler_info: ModelInfo,
scheduler_name: str,
) -> Scheduler:
scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(scheduler_name, SCHEDULER_MAP['ddim'])
orig_scheduler_info = context.services.model_manager.get_model(**scheduler_info.dict())
scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(
scheduler_name, SCHEDULER_MAP['ddim']
)
orig_scheduler_info = context.services.model_manager.get_model(
**scheduler_info.dict()
)
with orig_scheduler_info as orig_scheduler:
scheduler_config = orig_scheduler.config
if "_backup" in scheduler_config:
scheduler_config = scheduler_config["_backup"]
scheduler_config = {**scheduler_config, **scheduler_extra_config, "_backup": scheduler_config}
scheduler_config = {
**scheduler_config,
**scheduler_extra_config,
"_backup": scheduler_config,
}
scheduler = scheduler_class.from_config(scheduler_config)
# hack copied over from generate.py
if not hasattr(scheduler, 'uses_inpainting_model'):
scheduler.uses_inpainting_model = lambda: False
@ -124,17 +134,20 @@ class TextToLatentsInvocation(BaseInvocation):
"ui": {
"tags": ["latents"],
"type_hints": {
"model": "model",
"control": "control",
# "cfg_scale": "float",
"cfg_scale": "number"
"model": "model",
"control": "control",
# "cfg_scale": "float",
"cfg_scale": "number"
}
},
}
# TODO: pass this an emitter method or something? or a session for dispatching?
def dispatch_progress(
self, context: InvocationContext, source_node_id: str, intermediate_state: PipelineIntermediateState
self,
context: InvocationContext,
source_node_id: str,
intermediate_state: PipelineIntermediateState,
) -> None:
stable_diffusion_step_callback(
context=context,
@ -143,9 +156,17 @@ class TextToLatentsInvocation(BaseInvocation):
source_node_id=source_node_id,
)
def get_conditioning_data(self, context: InvocationContext, scheduler) -> ConditioningData:
c, extra_conditioning_info = context.services.latents.get(self.positive_conditioning.conditioning_name)
uc, _ = context.services.latents.get(self.negative_conditioning.conditioning_name)
def get_conditioning_data(
self,
context: InvocationContext,
scheduler,
) -> ConditioningData:
c, extra_conditioning_info = context.services.latents.get(
self.positive_conditioning.conditioning_name
)
uc, _ = context.services.latents.get(
self.negative_conditioning.conditioning_name
)
conditioning_data = ConditioningData(
unconditioned_embeddings=uc,
@ -153,10 +174,10 @@ class TextToLatentsInvocation(BaseInvocation):
guidance_scale=self.cfg_scale,
extra=extra_conditioning_info,
postprocessing_settings=PostprocessingSettings(
threshold=0.0,#threshold,
warmup=0.2,#warmup,
h_symmetry_time_pct=None,#h_symmetry_time_pct,
v_symmetry_time_pct=None#v_symmetry_time_pct,
threshold=0.0, # threshold,
warmup=0.2, # warmup,
h_symmetry_time_pct=None, # h_symmetry_time_pct,
v_symmetry_time_pct=None # v_symmetry_time_pct,
),
)
@ -164,31 +185,35 @@ class TextToLatentsInvocation(BaseInvocation):
scheduler,
# for ddim scheduler
eta=0.0, #ddim_eta
eta=0.0, # ddim_eta
# for ancestral and sde schedulers
generator=torch.Generator(device=uc.device).manual_seed(0),
)
return conditioning_data
def create_pipeline(self, unet, scheduler) -> StableDiffusionGeneratorPipeline:
def create_pipeline(
self,
unet,
scheduler,
) -> StableDiffusionGeneratorPipeline:
# TODO:
#configure_model_padding(
# configure_model_padding(
# unet,
# self.seamless,
# self.seamless_axes,
#)
# )
class FakeVae:
class FakeVaeConfig:
def __init__(self):
self.block_out_channels = [0]
def __init__(self):
self.config = FakeVae.FakeVaeConfig()
return StableDiffusionGeneratorPipeline(
vae=FakeVae(), # TODO: oh...
vae=FakeVae(), # TODO: oh...
text_encoder=None,
tokenizer=None,
unet=unet,
@ -198,13 +223,15 @@ class TextToLatentsInvocation(BaseInvocation):
requires_safety_checker=False,
precision="float16" if unet.dtype == torch.float16 else "float32",
)
def prep_control_data(
self,
context: InvocationContext,
model: StableDiffusionGeneratorPipeline, # really only need model for dtype and device
# really only need model for dtype and device
model: StableDiffusionGeneratorPipeline,
control_input: List[ControlField],
latents_shape: List[int],
exit_stack: ExitStack,
do_classifier_free_guidance: bool = True,
) -> List[ControlNetData]:
@ -230,23 +257,19 @@ class TextToLatentsInvocation(BaseInvocation):
control_data = []
control_models = []
for control_info in control_list:
# handle control models
if ("," in control_info.control_model):
control_model_split = control_info.control_model.split(",")
control_name = control_model_split[0]
control_subfolder = control_model_split[1]
print("Using HF model subfolders")
print(" control_name: ", control_name)
print(" control_subfolder: ", control_subfolder)
control_model = ControlNetModel.from_pretrained(control_name,
subfolder=control_subfolder,
torch_dtype=model.unet.dtype).to(model.device)
else:
control_model = ControlNetModel.from_pretrained(control_info.control_model,
torch_dtype=model.unet.dtype).to(model.device)
control_model = exit_stack.enter_context(
context.services.model_manager.get_model(
model_name=control_info.control_model.model_name,
model_type=ModelType.ControlNet,
base_model=control_info.control_model.base_model,
)
)
control_models.append(control_model)
control_image_field = control_info.image
input_image = context.services.images.get_pil_image(control_image_field.image_name)
input_image = context.services.images.get_pil_image(
control_image_field.image_name
)
# self.image.image_type, self.image.image_name
# FIXME: still need to test with different widths, heights, devices, dtypes
# and add in batch_size, num_images_per_prompt?
@ -263,58 +286,72 @@ class TextToLatentsInvocation(BaseInvocation):
dtype=control_model.dtype,
control_mode=control_info.control_mode,
)
control_item = ControlNetData(model=control_model,
image_tensor=control_image,
weight=control_info.control_weight,
begin_step_percent=control_info.begin_step_percent,
end_step_percent=control_info.end_step_percent,
control_mode=control_info.control_mode,
)
control_item = ControlNetData(
model=control_model, image_tensor=control_image,
weight=control_info.control_weight,
begin_step_percent=control_info.begin_step_percent,
end_step_percent=control_info.end_step_percent,
control_mode=control_info.control_mode,
)
control_data.append(control_item)
# MultiControlNetModel has been refactored out, just need list[ControlNetData]
return control_data
@torch.no_grad()
def invoke(self, context: InvocationContext) -> LatentsOutput:
noise = context.services.latents.get(self.noise.latents_name)
# Get the source node id (we are invoking the prepared node)
graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
graph_execution_state = context.services.graph_execution_manager.get(
context.graph_execution_state_id
)
source_node_id = graph_execution_state.prepared_source_mapping[self.id]
def step_callback(state: PipelineIntermediateState):
self.dispatch_progress(context, source_node_id, state)
unet_info = context.services.model_manager.get_model(**self.unet.unet.dict())
with unet_info as unet:
def _lora_loader():
for lora in self.unet.loras:
lora_info = context.services.model_manager.get_model(
**lora.dict(exclude={"weight"})
)
yield (lora_info.context.model, lora.weight)
del lora_info
return
unet_info = context.services.model_manager.get_model(
**self.unet.unet.dict()
)
with ExitStack() as exit_stack,\
ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
unet_info as unet:
scheduler = get_scheduler(
context=context,
scheduler_info=self.unet.scheduler,
scheduler_name=self.scheduler,
)
pipeline = self.create_pipeline(unet, scheduler)
conditioning_data = self.get_conditioning_data(context, scheduler)
loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.unet.loras]
control_data = self.prep_control_data(
model=pipeline, context=context, control_input=self.control,
latents_shape=noise.shape,
# do_classifier_free_guidance=(self.cfg_scale >= 1.0))
do_classifier_free_guidance=True,
exit_stack=exit_stack,
)
with ModelPatcher.apply_lora_unet(pipeline.unet, loras):
# TODO: Verify the noise is the right size
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
latents=torch.zeros_like(noise, dtype=torch_dtype(unet.device)),
noise=noise,
num_inference_steps=self.steps,
conditioning_data=conditioning_data,
control_data=control_data, # list[ControlNetData]
callback=step_callback,
)
# TODO: Verify the noise is the right size
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
latents=torch.zeros_like(noise, dtype=torch_dtype(unet.device)),
noise=noise,
num_inference_steps=self.steps,
conditioning_data=conditioning_data,
control_data=control_data, # list[ControlNetData]
callback=step_callback,
)
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
torch.cuda.empty_cache()
@ -323,14 +360,18 @@ class TextToLatentsInvocation(BaseInvocation):
context.services.latents.save(name, result_latents)
return build_latents_output(latents_name=name, latents=result_latents)
class LatentsToLatentsInvocation(TextToLatentsInvocation):
"""Generates latents using latents as base image."""
type: Literal["l2l"] = "l2l"
# Inputs
latents: Optional[LatentsField] = Field(description="The latents to use as a base image")
strength: float = Field(default=0.7, ge=0, le=1, description="The strength of the latents to use")
latents: Optional[LatentsField] = Field(
description="The latents to use as a base image")
strength: float = Field(
default=0.7, ge=0, le=1,
description="The strength of the latents to use")
# Schema customisation
class Config(InvocationConfig):
@ -345,22 +386,35 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
},
}
@torch.no_grad()
def invoke(self, context: InvocationContext) -> LatentsOutput:
noise = context.services.latents.get(self.noise.latents_name)
latent = context.services.latents.get(self.latents.latents_name)
# Get the source node id (we are invoking the prepared node)
graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
graph_execution_state = context.services.graph_execution_manager.get(
context.graph_execution_state_id
)
source_node_id = graph_execution_state.prepared_source_mapping[self.id]
def step_callback(state: PipelineIntermediateState):
self.dispatch_progress(context, source_node_id, state)
unet_info = context.services.model_manager.get_model(
**self.unet.unet.dict(),
)
def _lora_loader():
for lora in self.unet.loras:
lora_info = context.services.model_manager.get_model(
**lora.dict(exclude={"weight"})
)
yield (lora_info.context.model, lora.weight)
del lora_info
return
with unet_info as unet:
unet_info = context.services.model_manager.get_model(
**self.unet.unet.dict()
)
with ExitStack() as exit_stack,\
ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
unet_info as unet:
scheduler = get_scheduler(
context=context,
@ -370,12 +424,13 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
pipeline = self.create_pipeline(unet, scheduler)
conditioning_data = self.get_conditioning_data(context, scheduler)
control_data = self.prep_control_data(
model=pipeline, context=context, control_input=self.control,
latents_shape=noise.shape,
# do_classifier_free_guidance=(self.cfg_scale >= 1.0))
do_classifier_free_guidance=True,
exit_stack=exit_stack,
)
# TODO: Verify the noise is the right size
@ -389,18 +444,15 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
device=unet.device,
)
loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.unet.loras]
with ModelPatcher.apply_lora_unet(pipeline.unet, loras):
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
latents=initial_latents,
timesteps=timesteps,
noise=noise,
num_inference_steps=self.steps,
conditioning_data=conditioning_data,
control_data=control_data, # list[ControlNetData]
callback=step_callback
)
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
latents=initial_latents,
timesteps=timesteps,
noise=noise,
num_inference_steps=self.steps,
conditioning_data=conditioning_data,
control_data=control_data, # list[ControlNetData]
callback=step_callback
)
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
torch.cuda.empty_cache()
@ -417,9 +469,14 @@ class LatentsToImageInvocation(BaseInvocation):
type: Literal["l2i"] = "l2i"
# Inputs
latents: Optional[LatentsField] = Field(description="The latents to generate an image from")
latents: Optional[LatentsField] = Field(
description="The latents to generate an image from")
vae: VaeField = Field(default=None, description="Vae submodel")
tiled: bool = Field(default=False, description="Decode latents by overlaping tiles(less memory consumption)")
tiled: bool = Field(
default=False,
description="Decode latents by overlaping tiles(less memory consumption)")
metadata: Optional[CoreMetadata] = Field(default=None, description="Optional core metadata to be written to the image")
# Schema customisation
class Config(InvocationConfig):
@ -450,7 +507,7 @@ class LatentsToImageInvocation(BaseInvocation):
# copied from diffusers pipeline
latents = latents / vae.config.scaling_factor
image = vae.decode(latents, return_dict=False)[0]
image = (image / 2 + 0.5).clamp(0, 1) # denormalize
image = (image / 2 + 0.5).clamp(0, 1) # denormalize
# we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
np_image = image.cpu().permute(0, 2, 3, 1).float().numpy()
@ -464,7 +521,8 @@ class LatentsToImageInvocation(BaseInvocation):
image_category=ImageCategory.GENERAL,
node_id=self.id,
session_id=context.graph_execution_state_id,
is_intermediate=self.is_intermediate
is_intermediate=self.is_intermediate,
metadata=self.metadata.dict() if self.metadata else None,
)
return ImageOutput(
@ -473,9 +531,9 @@ class LatentsToImageInvocation(BaseInvocation):
height=image_dto.height,
)
LATENTS_INTERPOLATION_MODE = Literal[
"nearest", "linear", "bilinear", "bicubic", "trilinear", "area", "nearest-exact"
]
LATENTS_INTERPOLATION_MODE = Literal["nearest", "linear",
"bilinear", "bicubic", "trilinear", "area", "nearest-exact"]
class ResizeLatentsInvocation(BaseInvocation):
@ -484,20 +542,25 @@ class ResizeLatentsInvocation(BaseInvocation):
type: Literal["lresize"] = "lresize"
# Inputs
latents: Optional[LatentsField] = Field(description="The latents to resize")
width: int = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
height: int = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
mode: LATENTS_INTERPOLATION_MODE = Field(default="bilinear", description="The interpolation mode")
antialias: bool = Field(default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
latents: Optional[LatentsField] = Field(
description="The latents to resize")
width: Union[int, None] = Field(default=512,
ge=64, multiple_of=8, description="The width to resize to (px)")
height: Union[int, None] = Field(default=512,
ge=64, multiple_of=8, description="The height to resize to (px)")
mode: LATENTS_INTERPOLATION_MODE = Field(
default="bilinear", description="The interpolation mode")
antialias: bool = Field(
default=False,
description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
def invoke(self, context: InvocationContext) -> LatentsOutput:
latents = context.services.latents.get(self.latents.latents_name)
resized_latents = torch.nn.functional.interpolate(
latents,
size=(self.height // 8, self.width // 8),
mode=self.mode,
antialias=self.antialias if self.mode in ["bilinear", "bicubic"] else False,
latents, size=(self.height // 8, self.width // 8),
mode=self.mode, antialias=self.antialias
if self.mode in ["bilinear", "bicubic"] else False,
)
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
@ -515,20 +578,24 @@ class ScaleLatentsInvocation(BaseInvocation):
type: Literal["lscale"] = "lscale"
# Inputs
latents: Optional[LatentsField] = Field(description="The latents to scale")
scale_factor: float = Field(gt=0, description="The factor by which to scale the latents")
mode: LATENTS_INTERPOLATION_MODE = Field(default="bilinear", description="The interpolation mode")
antialias: bool = Field(default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
latents: Optional[LatentsField] = Field(
description="The latents to scale")
scale_factor: float = Field(
gt=0, description="The factor by which to scale the latents")
mode: LATENTS_INTERPOLATION_MODE = Field(
default="bilinear", description="The interpolation mode")
antialias: bool = Field(
default=False,
description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
def invoke(self, context: InvocationContext) -> LatentsOutput:
latents = context.services.latents.get(self.latents.latents_name)
# resizing
resized_latents = torch.nn.functional.interpolate(
latents,
scale_factor=self.scale_factor,
mode=self.mode,
antialias=self.antialias if self.mode in ["bilinear", "bicubic"] else False,
latents, scale_factor=self.scale_factor, mode=self.mode,
antialias=self.antialias
if self.mode in ["bilinear", "bicubic"] else False,
)
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
@ -546,9 +613,11 @@ class ImageToLatentsInvocation(BaseInvocation):
type: Literal["i2l"] = "i2l"
# Inputs
image: Union[ImageField, None] = Field(description="The image to encode")
image: Optional[ImageField] = Field(description="The image to encode")
vae: VaeField = Field(default=None, description="Vae submodel")
tiled: bool = Field(default=False, description="Encode latents by overlaping tiles(less memory consumption)")
tiled: bool = Field(
default=False,
description="Encode latents by overlaping tiles(less memory consumption)")
# Schema customisation
class Config(InvocationConfig):

124
invokeai/app/invocations/metadata.py Normal file
View File

@ -0,0 +1,124 @@
from typing import Literal, Optional, Union
from pydantic import BaseModel, Field
from invokeai.app.invocations.baseinvocation import (BaseInvocation,
BaseInvocationOutput,
InvocationContext)
from invokeai.app.invocations.controlnet_image_processors import ControlField
from invokeai.app.invocations.model import (LoRAModelField, MainModelField,
VAEModelField)
class LoRAMetadataField(BaseModel):
"""LoRA metadata for an image generated in InvokeAI."""
lora: LoRAModelField = Field(description="The LoRA model")
weight: float = Field(description="The weight of the LoRA model")
class CoreMetadata(BaseModel):
"""Core generation metadata for an image generated in InvokeAI."""
generation_mode: str = Field(description="The generation mode that output this image",)
positive_prompt: str = Field(description="The positive prompt parameter")
negative_prompt: str = Field(description="The negative prompt parameter")
width: int = Field(description="The width parameter")
height: int = Field(description="The height parameter")
seed: int = Field(description="The seed used for noise generation")
rand_device: str = Field(description="The device used for random number generation")
cfg_scale: float = Field(description="The classifier-free guidance scale parameter")
steps: int = Field(description="The number of steps used for inference")
scheduler: str = Field(description="The scheduler used for inference")
clip_skip: int = Field(description="The number of skipped CLIP layers",)
model: MainModelField = Field(description="The main model used for inference")
controlnets: list[ControlField]= Field(description="The ControlNets used for inference")
loras: list[LoRAMetadataField] = Field(description="The LoRAs used for inference")
strength: Union[float, None] = Field(
default=None,
description="The strength used for latents-to-latents",
)
init_image: Union[str, None] = Field(
default=None, description="The name of the initial image"
)
vae: Union[VAEModelField, None] = Field(
default=None,
description="The VAE used for decoding, if the main model's default was not used",
)
class ImageMetadata(BaseModel):
"""An image's generation metadata"""
metadata: Optional[dict] = Field(
default=None,
description="The image's core metadata, if it was created in the Linear or Canvas UI",
)
graph: Optional[dict] = Field(
default=None, description="The graph that created the image"
)
class MetadataAccumulatorOutput(BaseInvocationOutput):
"""The output of the MetadataAccumulator node"""
type: Literal["metadata_accumulator_output"] = "metadata_accumulator_output"
metadata: CoreMetadata = Field(description="The core metadata for the image")
class MetadataAccumulatorInvocation(BaseInvocation):
"""Outputs a Core Metadata Object"""
type: Literal["metadata_accumulator"] = "metadata_accumulator"
generation_mode: str = Field(description="The generation mode that output this image",)
positive_prompt: str = Field(description="The positive prompt parameter")
negative_prompt: str = Field(description="The negative prompt parameter")
width: int = Field(description="The width parameter")
height: int = Field(description="The height parameter")
seed: int = Field(description="The seed used for noise generation")
rand_device: str = Field(description="The device used for random number generation")
cfg_scale: float = Field(description="The classifier-free guidance scale parameter")
steps: int = Field(description="The number of steps used for inference")
scheduler: str = Field(description="The scheduler used for inference")
clip_skip: int = Field(description="The number of skipped CLIP layers",)
model: MainModelField = Field(description="The main model used for inference")
controlnets: list[ControlField]= Field(description="The ControlNets used for inference")
loras: list[LoRAMetadataField] = Field(description="The LoRAs used for inference")
strength: Union[float, None] = Field(
default=None,
description="The strength used for latents-to-latents",
)
init_image: Union[str, None] = Field(
default=None, description="The name of the initial image"
)
vae: Union[VAEModelField, None] = Field(
default=None,
description="The VAE used for decoding, if the main model's default was not used",
)
def invoke(self, context: InvocationContext) -> MetadataAccumulatorOutput:
"""Collects and outputs a CoreMetadata object"""
return MetadataAccumulatorOutput(
metadata=CoreMetadata(
generation_mode=self.generation_mode,
positive_prompt=self.positive_prompt,
negative_prompt=self.negative_prompt,
width=self.width,
height=self.height,
seed=self.seed,
rand_device=self.rand_device,
cfg_scale=self.cfg_scale,
steps=self.steps,
scheduler=self.scheduler,
model=self.model,
strength=self.strength,
init_image=self.init_image,
vae=self.vae,
controlnets=self.controlnets,
loras=self.loras,
clip_skip=self.clip_skip,
)
)

153
invokeai/app/invocations/model.py
View File

@ -1,31 +1,39 @@
from typing import Literal, Optional, Union, List
from pydantic import BaseModel, Field
import copy
from typing import List, Literal, Optional, Union
from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext, InvocationConfig
from pydantic import BaseModel, Field
from ...backend.util.devices import choose_torch_device, torch_dtype
from ...backend.model_management import BaseModelType, ModelType, SubModelType
from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
InvocationConfig, InvocationContext)
class ModelInfo(BaseModel):
model_name: str = Field(description="Info to load submodel")
base_model: BaseModelType = Field(description="Base model")
model_type: ModelType = Field(description="Info to load submodel")
submodel: Optional[SubModelType] = Field(description="Info to load submodel")
submodel: Optional[SubModelType] = Field(
default=None, description="Info to load submodel"
)
class LoraInfo(ModelInfo):
weight: float = Field(description="Lora's weight which to use when apply to model")
class UNetField(BaseModel):
unet: ModelInfo = Field(description="Info to load unet submodel")
scheduler: ModelInfo = Field(description="Info to load scheduler submodel")
loras: List[LoraInfo] = Field(description="Loras to apply on model loading")
class ClipField(BaseModel):
tokenizer: ModelInfo = Field(description="Info to load tokenizer submodel")
text_encoder: ModelInfo = Field(description="Info to load text_encoder submodel")
skipped_layers: int = Field(description="Number of skipped layers in text_encoder")
loras: List[LoraInfo] = Field(description="Loras to apply on model loading")
class VaeField(BaseModel):
# TODO: better naming?
vae: ModelInfo = Field(description="Info to load vae submodel")
@ -34,43 +42,48 @@ class VaeField(BaseModel):
class ModelLoaderOutput(BaseInvocationOutput):
"""Model loader output"""
#fmt: off
# fmt: off
type: Literal["model_loader_output"] = "model_loader_output"
unet: UNetField = Field(default=None, description="UNet submodel")
clip: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
vae: VaeField = Field(default=None, description="Vae submodel")
#fmt: on
# fmt: on
class PipelineModelField(BaseModel):
"""Pipeline model field"""
class MainModelField(BaseModel):
"""Main model field"""
model_name: str = Field(description="Name of the model")
base_model: BaseModelType = Field(description="Base model")
class PipelineModelLoaderInvocation(BaseInvocation):
"""Loads a pipeline model, outputting its submodels."""
class LoRAModelField(BaseModel):
"""LoRA model field"""
type: Literal["pipeline_model_loader"] = "pipeline_model_loader"
model_name: str = Field(description="Name of the LoRA model")
base_model: BaseModelType = Field(description="Base model")
model: PipelineModelField = Field(description="The model to load")
class MainModelLoaderInvocation(BaseInvocation):
"""Loads a main model, outputting its submodels."""
type: Literal["main_model_loader"] = "main_model_loader"
model: MainModelField = Field(description="The model to load")
# TODO: precision?
# Schema customisation
class Config(InvocationConfig):
schema_extra = {
"ui": {
"title": "Model Loader",
"tags": ["model", "loader"],
"type_hints": {
"model": "model"
}
"type_hints": {"model": "model"},
},
}
def invoke(self, context: InvocationContext) -> ModelLoaderOutput:
base_model = self.model.base_model
model_name = self.model.model_name
model_type = ModelType.Main
@ -112,7 +125,6 @@ class PipelineModelLoaderInvocation(BaseInvocation):
)
"""
return ModelLoaderOutput(
unet=UNetField(
unet=ModelInfo(
@ -143,6 +155,7 @@ class PipelineModelLoaderInvocation(BaseInvocation):
submodel=SubModelType.TextEncoder,
),
loras=[],
skipped_layers=0,
),
vae=VaeField(
vae=ModelInfo(
@ -151,47 +164,66 @@ class PipelineModelLoaderInvocation(BaseInvocation):
model_type=model_type,
submodel=SubModelType.Vae,
),
)
),
)
class LoraLoaderOutput(BaseInvocationOutput):
"""Model loader output"""
#fmt: off
# fmt: off
type: Literal["lora_loader_output"] = "lora_loader_output"
unet: Optional[UNetField] = Field(default=None, description="UNet submodel")
clip: Optional[ClipField] = Field(default=None, description="Tokenizer and text_encoder submodels")
#fmt: on
# fmt: on
class LoraLoaderInvocation(BaseInvocation):
"""Apply selected lora to unet and text_encoder."""
type: Literal["lora_loader"] = "lora_loader"
lora_name: str = Field(description="Lora model name")
lora: Union[LoRAModelField, None] = Field(
default=None, description="Lora model name"
)
weight: float = Field(default=0.75, description="With what weight to apply lora")
unet: Optional[UNetField] = Field(description="UNet model for applying lora")
clip: Optional[ClipField] = Field(description="Clip model for applying lora")
def invoke(self, context: InvocationContext) -> LoraLoaderOutput:
class Config(InvocationConfig):
schema_extra = {
"ui": {
"title": "Lora Loader",
"tags": ["lora", "loader"],
"type_hints": {"lora": "lora_model"},
},
}
# TODO: ui rewrite
base_model = BaseModelType.StableDiffusion1
def invoke(self, context: InvocationContext) -> LoraLoaderOutput:
if self.lora is None:
raise Exception("No LoRA provided")
base_model = self.lora.base_model
lora_name = self.lora.model_name
if not context.services.model_manager.model_exists(
base_model=base_model,
model_name=self.lora_name,
model_name=lora_name,
model_type=ModelType.Lora,
):
raise Exception(f"Unkown lora name: {self.lora_name}!")
raise Exception(f"Unkown lora name: {lora_name}!")
if self.unet is not None and any(lora.model_name == self.lora_name for lora in self.unet.loras):
raise Exception(f"Lora \"{self.lora_name}\" already applied to unet")
if self.unet is not None and any(
lora.model_name == lora_name for lora in self.unet.loras
):
raise Exception(f'Lora "{lora_name}" already applied to unet')
if self.clip is not None and any(lora.model_name == self.lora_name for lora in self.clip.loras):
raise Exception(f"Lora \"{self.lora_name}\" already applied to clip")
if self.clip is not None and any(
lora.model_name == lora_name for lora in self.clip.loras
):
raise Exception(f'Lora "{lora_name}" already applied to clip')
output = LoraLoaderOutput()
@ -200,7 +232,7 @@ class LoraLoaderInvocation(BaseInvocation):
output.unet.loras.append(
LoraInfo(
base_model=base_model,
model_name=self.lora_name,
model_name=lora_name,
model_type=ModelType.Lora,
submodel=None,
weight=self.weight,
@ -212,7 +244,7 @@ class LoraLoaderInvocation(BaseInvocation):
output.clip.loras.append(
LoraInfo(
base_model=base_model,
model_name=self.lora_name,
model_name=lora_name,
model_type=ModelType.Lora,
submodel=None,
weight=self.weight,
@ -221,3 +253,58 @@ class LoraLoaderInvocation(BaseInvocation):
return output
class VAEModelField(BaseModel):
"""Vae model field"""
model_name: str = Field(description="Name of the model")
base_model: BaseModelType = Field(description="Base model")
class VaeLoaderOutput(BaseInvocationOutput):
"""Model loader output"""
# fmt: off
type: Literal["vae_loader_output"] = "vae_loader_output"
vae: VaeField = Field(default=None, description="Vae model")
# fmt: on
class VaeLoaderInvocation(BaseInvocation):
"""Loads a VAE model, outputting a VaeLoaderOutput"""
type: Literal["vae_loader"] = "vae_loader"
vae_model: VAEModelField = Field(description="The VAE to load")
# Schema customisation
class Config(InvocationConfig):
schema_extra = {
"ui": {
"title": "VAE Loader",
"tags": ["vae", "loader"],
"type_hints": {"vae_model": "vae_model"},
},
}
def invoke(self, context: InvocationContext) -> VaeLoaderOutput:
base_model = self.vae_model.base_model
model_name = self.vae_model.model_name
model_type = ModelType.Vae
if not context.services.model_manager.model_exists(
base_model=base_model,
model_name=model_name,
model_type=model_type,
):
raise Exception(f"Unkown vae name: {model_name}!")
return VaeLoaderOutput(
vae=VaeField(
vae=ModelInfo(
model_name=model_name,
base_model=base_model,
model_type=model_type,
)
)
)

2
invokeai/app/invocations/noise.py
View File

@ -32,7 +32,7 @@ def get_noise(
perlin: float = 0.0,
):
"""Generate noise for a given image size."""
noise_device_type = "cpu" if (use_cpu or device.type == "mps") else device.type
noise_device_type = "cpu" if use_cpu else device.type
# limit noise to only the diffusion image channels, not the mask channels
input_channels = min(latent_channels, 4)

44
invokeai/app/invocations/prompt.py
View File

@ -1,6 +1,8 @@
from typing import Literal
from os.path import exists
from typing import Literal, Optional
from pydantic.fields import Field
import numpy as np
from pydantic import Field, validator
from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext
from dynamicprompts.generators import RandomPromptGenerator, CombinatorialPromptGenerator
@ -55,3 +57,41 @@ class DynamicPromptInvocation(BaseInvocation):
prompts = generator.generate(self.prompt, num_images=self.max_prompts)
return PromptCollectionOutput(prompt_collection=prompts, count=len(prompts))
class PromptsFromFileInvocation(BaseInvocation):
'''Loads prompts from a text file'''
# fmt: off
type: Literal['prompt_from_file'] = 'prompt_from_file'
# Inputs
file_path: str = Field(description="Path to prompt text file")
pre_prompt: Optional[str] = Field(description="String to prepend to each prompt")
post_prompt: Optional[str] = Field(description="String to append to each prompt")
start_line: int = Field(default=1, ge=1, description="Line in the file to start start from")
max_prompts: int = Field(default=1, ge=0, description="Max lines to read from file (0=all)")
#fmt: on
@validator("file_path")
def file_path_exists(cls, v):
if not exists(v):
raise ValueError(FileNotFoundError)
return v
def promptsFromFile(self, file_path: str, pre_prompt: str, post_prompt: str, start_line: int, max_prompts: int):
prompts = []
start_line -= 1
end_line = start_line + max_prompts
if max_prompts <= 0:
end_line = np.iinfo(np.int32).max
with open(file_path) as f:
for i, line in enumerate(f):
if i >= start_line and i < end_line:
prompts.append((pre_prompt or '') + line.strip() + (post_prompt or ''))
if i >= end_line:
break
return prompts
def invoke(self, context: InvocationContext) -> PromptCollectionOutput:
prompts = self.promptsFromFile(self.file_path, self.pre_prompt, self.post_prompt, self.start_line, self.max_prompts)
return PromptCollectionOutput(prompt_collection=prompts, count=len(prompts))

55
invokeai/app/invocations/reconstruct.py
View File

@ -1,55 +0,0 @@
from typing import Literal, Union
from pydantic import Field
from invokeai.app.models.image import ImageCategory, ImageField, ResourceOrigin
from .baseinvocation import BaseInvocation, InvocationContext, InvocationConfig
from .image import ImageOutput
class RestoreFaceInvocation(BaseInvocation):
"""Restores faces in an image."""
# fmt: off
type: Literal["restore_face"] = "restore_face"
# Inputs
image: Union[ImageField, None] = Field(description="The input image")
strength: float = Field(default=0.75, gt=0, le=1, description="The strength of the restoration" )
# fmt: on
# Schema customisation
class Config(InvocationConfig):
schema_extra = {
"ui": {
"tags": ["restoration", "image"],
},
}
def invoke(self, context: InvocationContext) -> ImageOutput:
image = context.services.images.get_pil_image(self.image.image_name)
results = context.services.restoration.upscale_and_reconstruct(
image_list=[[image, 0]],
upscale=None,
strength=self.strength, # GFPGAN strength
save_original=False,
image_callback=None,
)
# Results are image and seed, unwrap for now
# TODO: can this return multiple results?
image_dto = context.services.images.create(
image=results[0][0],
image_origin=ResourceOrigin.INTERNAL,
image_category=ImageCategory.GENERAL,
node_id=self.id,
session_id=context.graph_execution_state_id,
is_intermediate=self.is_intermediate,
)
return ImageOutput(
image=ImageField(image_name=image_dto.image_name),
width=image_dto.width,
height=image_dto.height,
)

124
invokeai/app/invocations/upscale.py
View File

@ -1,48 +1,112 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
from typing import Literal, Union
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654) & the InvokeAI Team
from pathlib import Path, PosixPath
from typing import Literal, Union, cast
import cv2 as cv
import numpy as np
from basicsr.archs.rrdbnet_arch import RRDBNet
from PIL import Image
from pydantic import Field
from realesrgan import RealESRGANer
from invokeai.app.models.image import ImageCategory, ImageField, ResourceOrigin
from .baseinvocation import BaseInvocation, InvocationContext, InvocationConfig
from .baseinvocation import BaseInvocation, InvocationContext
from .image import ImageOutput
# TODO: Populate this from disk?
# TODO: Use model manager to load?
REALESRGAN_MODELS = Literal[
"RealESRGAN_x4plus.pth",
"RealESRGAN_x4plus_anime_6B.pth",
"ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
]
class UpscaleInvocation(BaseInvocation):
"""Upscales an image."""
# fmt: off
type: Literal["upscale"] = "upscale"
class RealESRGANInvocation(BaseInvocation):
"""Upscales an image using RealESRGAN."""
# Inputs
image: Union[ImageField, None] = Field(description="The input image", default=None)
strength: float = Field(default=0.75, gt=0, le=1, description="The strength")
level: Literal[2, 4] = Field(default=2, description="The upscale level")
# fmt: on
# Schema customisation
class Config(InvocationConfig):
schema_extra = {
"ui": {
"tags": ["upscaling", "image"],
},
}
type: Literal["realesrgan"] = "realesrgan"
image: Union[ImageField, None] = Field(default=None, description="The input image")
model_name: REALESRGAN_MODELS = Field(
default="RealESRGAN_x4plus.pth", description="The Real-ESRGAN model to use"
)
def invoke(self, context: InvocationContext) -> ImageOutput:
image = context.services.images.get_pil_image(self.image.image_name)
results = context.services.restoration.upscale_and_reconstruct(
image_list=[[image, 0]],
upscale=(self.level, self.strength),
strength=0.0, # GFPGAN strength
save_original=False,
image_callback=None,
models_path = context.services.configuration.models_path
rrdbnet_model = None
netscale = None
esrgan_model_path = None
if self.model_name in [
"RealESRGAN_x4plus.pth",
"ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
]:
# x4 RRDBNet model
rrdbnet_model = RRDBNet(
num_in_ch=3,
num_out_ch=3,
num_feat=64,
num_block=23,
num_grow_ch=32,
scale=4,
)
netscale = 4
elif self.model_name in ["RealESRGAN_x4plus_anime_6B.pth"]:
# x4 RRDBNet model, 6 blocks
rrdbnet_model = RRDBNet(
num_in_ch=3,
num_out_ch=3,
num_feat=64,
num_block=6, # 6 blocks
num_grow_ch=32,
scale=4,
)
netscale = 4
# TODO: add x2 models handling?
# elif self.model_name in ["RealESRGAN_x2plus"]:
# # x2 RRDBNet model
# model = RRDBNet(
# num_in_ch=3,
# num_out_ch=3,
# num_feat=64,
# num_block=23,
# num_grow_ch=32,
# scale=2,
# )
# model_path = Path()
# netscale = 2
else:
msg = f"Invalid RealESRGAN model: {self.model_name}"
context.services.logger.error(msg)
raise ValueError(msg)
esrgan_model_path = Path(f"core/upscaling/realesrgan/{self.model_name}")
upsampler = RealESRGANer(
scale=netscale,
model_path=str(models_path / esrgan_model_path),
model=rrdbnet_model,
half=False,
)
# Results are image and seed, unwrap for now
# TODO: can this return multiple results?
# prepare image - Real-ESRGAN uses cv2 internally, and cv2 uses BGR vs RGB for PIL
cv_image = cv.cvtColor(np.array(image.convert("RGB")), cv.COLOR_RGB2BGR)
# We can pass an `outscale` value here, but it just resizes the image by that factor after
# upscaling, so it's kinda pointless for our purposes. If you want something other than 4x
# upscaling, you'll need to add a resize node after this one.
upscaled_image, img_mode = upsampler.enhance(cv_image)
# back to PIL
pil_image = Image.fromarray(
cv.cvtColor(upscaled_image, cv.COLOR_BGR2RGB)
).convert("RGBA")
image_dto = context.services.images.create(
image=results[0][0],
image=pil_image,
image_origin=ResourceOrigin.INTERNAL,
image_category=ImageCategory.GENERAL,
node_id=self.id,

93
invokeai/app/models/metadata.py
View File

@ -1,93 +0,0 @@
from typing import Optional, Union, List
from pydantic import BaseModel, Extra, Field, StrictFloat, StrictInt, StrictStr
class ImageMetadata(BaseModel):
"""
Core generation metadata for an image/tensor generated in InvokeAI.
Also includes any metadata from the image's PNG tEXt chunks.
Generated by traversing the execution graph, collecting the parameters of the nearest ancestors
of a given node.
Full metadata may be accessed by querying for the session in the `graph_executions` table.
"""
class Config:
extra = Extra.allow
"""
This lets the ImageMetadata class accept arbitrary additional fields. The CoreMetadataService
won't add any fields that are not already defined, but other a different metadata service
implementation might.
"""
type: Optional[StrictStr] = Field(
default=None,
description="The type of the ancestor node of the image output node.",
)
"""The type of the ancestor node of the image output node."""
positive_conditioning: Optional[StrictStr] = Field(
default=None, description="The positive conditioning."
)
"""The positive conditioning"""
negative_conditioning: Optional[StrictStr] = Field(
default=None, description="The negative conditioning."
)
"""The negative conditioning"""
width: Optional[StrictInt] = Field(
default=None, description="Width of the image/latents in pixels."
)
"""Width of the image/latents in pixels"""
height: Optional[StrictInt] = Field(
default=None, description="Height of the image/latents in pixels."
)
"""Height of the image/latents in pixels"""
seed: Optional[StrictInt] = Field(
default=None, description="The seed used for noise generation."
)
"""The seed used for noise generation"""
# cfg_scale: Optional[StrictFloat] = Field(
# cfg_scale: Union[float, list[float]] = Field(
cfg_scale: Union[StrictFloat, List[StrictFloat]] = Field(
default=None, description="The classifier-free guidance scale."
)
"""The classifier-free guidance scale"""
steps: Optional[StrictInt] = Field(
default=None, description="The number of steps used for inference."
)
"""The number of steps used for inference"""
scheduler: Optional[StrictStr] = Field(
default=None, description="The scheduler used for inference."
)
"""The scheduler used for inference"""
model: Optional[StrictStr] = Field(
default=None, description="The model used for inference."
)
"""The model used for inference"""
strength: Optional[StrictFloat] = Field(
default=None,
description="The strength used for image-to-image/latents-to-latents.",
)
"""The strength used for image-to-image/latents-to-latents."""
latents: Optional[StrictStr] = Field(
default=None, description="The ID of the initial latents."
)
"""The ID of the initial latents"""
vae: Optional[StrictStr] = Field(
default=None, description="The VAE used for decoding."
)
"""The VAE used for decoding"""
unet: Optional[StrictStr] = Field(
default=None, description="The UNet used dor inference."
)
"""The UNet used dor inference"""
clip: Optional[StrictStr] = Field(
default=None, description="The CLIP Encoder used for conditioning."
)
"""The CLIP Encoder used for conditioning"""
extra: Optional[StrictStr] = Field(
default=None,
description="Uploaded image metadata, extracted from the PNG tEXt chunk.",
)
"""Uploaded image metadata, extracted from the PNG tEXt chunk."""

7
invokeai/app/services/board_image_record_storage.py
View File

@ -1,8 +1,7 @@
from abc import ABC, abstractmethod
import sqlite3
import threading
from typing import Union, cast
from invokeai.app.services.board_record_storage import BoardRecord
from typing import Optional, cast
from invokeai.app.services.image_record_storage import OffsetPaginatedResults
from invokeai.app.services.models.image_record import (
@ -44,7 +43,7 @@ class BoardImageRecordStorageBase(ABC):
def get_board_for_image(
self,
image_name: str,
) -> Union[str, None]:
) -> Optional[str]:
"""Gets an image's board id, if it has one."""
pass
@ -215,7 +214,7 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
def get_board_for_image(
self,
image_name: str,
) -> Union[str, None]:
) -> Optional[str]:
try:
self._lock.acquire()
self._cursor.execute(

8
invokeai/app/services/board_images.py
View File

@ -1,6 +1,6 @@
from abc import ABC, abstractmethod
from logging import Logger
from typing import List, Union
from typing import List, Union, Optional
from invokeai.app.services.board_image_record_storage import BoardImageRecordStorageBase
from invokeai.app.services.board_record_storage import (
BoardRecord,
@ -49,7 +49,7 @@ class BoardImagesServiceABC(ABC):
def get_board_for_image(
self,
image_name: str,
) -> Union[str, None]:
) -> Optional[str]:
"""Gets an image's board id, if it has one."""
pass
@ -126,13 +126,13 @@ class BoardImagesService(BoardImagesServiceABC):
def get_board_for_image(
self,
image_name: str,
) -> Union[str, None]:
) -> Optional[str]:
board_id = self._services.board_image_records.get_board_for_image(image_name)
return board_id
def board_record_to_dto(
board_record: BoardRecord, cover_image_name: str | None, image_count: int
board_record: BoardRecord, cover_image_name: Optional[str], image_count: int
) -> BoardDTO:
"""Converts a board record to a board DTO."""
return BoardDTO(

50
invokeai/app/services/config.py
View File

@ -23,7 +23,8 @@ InvokeAI:
xformers_enabled: false
sequential_guidance: false
precision: float16
max_loaded_models: 4
max_cache_size: 6
max_vram_cache_size: 2.7
always_use_cpu: false
free_gpu_mem: false
Features:
@ -168,9 +169,10 @@ from argparse import ArgumentParser
from omegaconf import OmegaConf, DictConfig
from pathlib import Path
from pydantic import BaseSettings, Field, parse_obj_as
from typing import ClassVar, Dict, List, Literal, Union, get_origin, get_type_hints, get_args
from typing import ClassVar, Dict, List, Set, Literal, Union, get_origin, get_type_hints, get_args
INIT_FILE = Path('invokeai.yaml')
MODEL_CORE = Path('models/core')
DB_FILE = Path('invokeai.db')
LEGACY_INIT_FILE = Path('invokeai.init')
@ -198,7 +200,7 @@ class InvokeAISettings(BaseSettings):
type = get_args(get_type_hints(cls)['type'])[0]
field_dict = dict({type:dict()})
for name,field in self.__fields__.items():
if name in cls._excluded():
if name in cls._excluded_from_yaml():
continue
category = field.field_info.extra.get("category") or "Uncategorized"
value = getattr(self,name)
@ -228,10 +230,10 @@ class InvokeAISettings(BaseSettings):
upcase_environ = dict()
for key,value in os.environ.items():
upcase_environ[key.upper()] = value
fields = cls.__fields__
cls.argparse_groups = {}
for name, field in fields.items():
if name not in cls._excluded():
current_default = field.default
@ -269,7 +271,13 @@ class InvokeAISettings(BaseSettings):
@classmethod
def _excluded(self)->List[str]:
# internal fields that shouldn't be exposed as command line options
return ['type','initconf']
@classmethod
def _excluded_from_yaml(self)->List[str]:
# combination of deprecated parameters and internal ones that shouldn't be exposed as invokeai.yaml options
return ['type','initconf', 'gpu_mem_reserved', 'max_loaded_models', 'version', 'from_file', 'model', 'restore']
class Config:
env_file_encoding = 'utf-8'
@ -324,16 +332,11 @@ class InvokeAISettings(BaseSettings):
help=field.field_info.description,
)
def _find_root()->Path:
venv = Path(os.environ.get("VIRTUAL_ENV") or ".")
if os.environ.get("INVOKEAI_ROOT"):
root = Path(os.environ.get("INVOKEAI_ROOT")).resolve()
elif (
os.environ.get("VIRTUAL_ENV")
and (Path(os.environ.get("VIRTUAL_ENV"), "..", INIT_FILE).exists()
or
Path(os.environ.get("VIRTUAL_ENV"), "..", LEGACY_INIT_FILE).exists()
)
):
root = Path(os.environ.get("VIRTUAL_ENV"), "..").resolve()
elif any([(venv.parent/x).exists() for x in [INIT_FILE, LEGACY_INIT_FILE, MODEL_CORE]]):
root = (venv.parent).resolve()
else:
root = Path("~/invokeai").expanduser().resolve()
return root
@ -348,7 +351,7 @@ setting environment variables INVOKEAI_<setting>.
'''
singleton_config: ClassVar[InvokeAIAppConfig] = None
singleton_init: ClassVar[Dict] = None
#fmt: off
type: Literal["InvokeAI"] = "InvokeAI"
host : str = Field(default="127.0.0.1", description="IP address to bind to", category='Web Server')
@ -363,11 +366,14 @@ setting environment variables INVOKEAI_<setting>.
log_tokenization : bool = Field(default=False, description="Enable logging of parsed prompt tokens.", category='Features')
nsfw_checker : bool = Field(default=True, description="Enable/disable the NSFW checker", category='Features')
patchmatch : bool = Field(default=True, description="Enable/disable patchmatch inpaint code", category='Features')
restore : bool = Field(default=True, description="Enable/disable face restoration code", category='Features')
restore : bool = Field(default=True, description="Enable/disable face restoration code (DEPRECATED)", category='DEPRECATED')
always_use_cpu : bool = Field(default=False, description="If true, use the CPU for rendering even if a GPU is available.", category='Memory/Performance')
free_gpu_mem : bool = Field(default=False, description="If true, purge model from GPU after each generation.", category='Memory/Performance')
max_loaded_models : int = Field(default=3, gt=0, description="Maximum number of models to keep in memory for rapid switching", category='Memory/Performance')
max_loaded_models : int = Field(default=3, gt=0, description="(DEPRECATED: use max_cache_size) Maximum number of models to keep in memory for rapid switching", category='DEPRECATED')
max_cache_size : float = Field(default=6.0, gt=0, description="Maximum memory amount used by model cache for rapid switching", category='Memory/Performance')
max_vram_cache_size : float = Field(default=2.75, ge=0, description="Amount of VRAM reserved for model storage", category='Memory/Performance')
gpu_mem_reserved : float = Field(default=2.75, ge=0, description="DEPRECATED: use max_vram_cache_size. Amount of VRAM reserved for model storage", category='DEPRECATED')
precision : Literal[tuple(['auto','float16','float32','autocast'])] = Field(default='float16',description='Floating point precision', category='Memory/Performance')
sequential_guidance : bool = Field(default=False, description="Whether to calculate guidance in serial instead of in parallel, lowering memory requirements", category='Memory/Performance')
xformers_enabled : bool = Field(default=True, description="Enable/disable memory-efficient attention", category='Memory/Performance')
@ -385,18 +391,20 @@ setting environment variables INVOKEAI_<setting>.
outdir : Path = Field(default='outputs', description='Default folder for output images', category='Paths')
from_file : Path = Field(default=None, description='Take command input from the indicated file (command-line client only)', category='Paths')
use_memory_db : bool = Field(default=False, description='Use in-memory database for storing image metadata', category='Paths')
model : str = Field(default='stable-diffusion-1.5', description='Initial model name', category='Models')
log_handlers : List[str] = Field(default=["console"], description='Log handler. Valid options are "console", "file=<path>", "syslog=path|address:host:port", "http=<url>"', category="Logging")
# note - would be better to read the log_format values from logging.py, but this creates circular dependencies issues
log_format : Literal[tuple(['plain','color','syslog','legacy'])] = Field(default="color", description='Log format. Use "plain" for text-only, "color" for colorized output, "legacy" for 2.3-style logging and "syslog" for syslog-style', category="Logging")
log_level : Literal[tuple(["debug","info","warning","error","critical"])] = Field(default="debug", description="Emit logging messages at this level or higher", category="Logging")
version : bool = Field(default=False, description="Show InvokeAI version and exit", category="Other")
#fmt: on
def parse_args(self, argv: List[str]=None, conf: DictConfig = None, clobber=False):
'''
Update settings with contents of init file, environment, and
Update settings with contents of init file, environment, and
command-line settings.
:param conf: alternate Omegaconf dictionary object
:param argv: aternate sys.argv list
@ -411,7 +419,7 @@ setting environment variables INVOKEAI_<setting>.
except:
pass
InvokeAISettings.initconf = conf
# parse args again in order to pick up settings in configuration file
super().parse_args(argv)
@ -431,7 +439,7 @@ setting environment variables INVOKEAI_<setting>.
cls.singleton_config = cls(**kwargs)
cls.singleton_init = kwargs
return cls.singleton_config
@property
def root_path(self)->Path:
'''

5
invokeai/app/services/events.py
View File

@ -1,10 +1,9 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
from typing import Any
from typing import Any, Optional
from invokeai.app.models.image import ProgressImage
from invokeai.app.util.misc import get_timestamp
from invokeai.app.services.model_manager_service import BaseModelType, ModelType, SubModelType, ModelInfo
from invokeai.app.models.exceptions import CanceledException
class EventServiceBase:
session_event: str = "session_event"
@ -28,7 +27,7 @@ class EventServiceBase:
graph_execution_state_id: str,
node: dict,
source_node_id: str,
progress_image: ProgressImage | None,
progress_image: Optional[ProgressImage],
step: int,
total_steps: int,
) -> None:

7
invokeai/app/services/graph.py
View File

@ -3,7 +3,6 @@
import copy
import itertools
import uuid
from types import NoneType
from typing import (
Annotated,
Any,
@ -26,6 +25,8 @@ from ..invocations.baseinvocation import (
InvocationContext,
)
# in 3.10 this would be "from types import NoneType"
NoneType = type(None)
class EdgeConnection(BaseModel):
node_id: str = Field(description="The id of the node for this edge connection")
@ -60,8 +61,6 @@ def get_input_field(node: BaseInvocation, field: str) -> Any:
node_input_field = node_inputs.get(field) or None
return node_input_field
from typing import Optional, Union, List, get_args
def is_union_subtype(t1, t2):
t1_args = get_args(t1)
t2_args = get_args(t2)
@ -846,7 +845,7 @@ class GraphExecutionState(BaseModel):
]
}
def next(self) -> BaseInvocation | None:
def next(self) -> Optional[BaseInvocation]:
"""Gets the next node ready to execute."""
# TODO: enable multiple nodes to execute simultaneously by tracking currently executing nodes

30
invokeai/app/services/image_file_storage.py
View File

@ -1,15 +1,14 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654) and the InvokeAI Team
import json
from abc import ABC, abstractmethod
from pathlib import Path
from queue import Queue
from typing import Dict, Optional
from typing import Dict, Optional, Union
from PIL.Image import Image as PILImageType
from PIL import Image, PngImagePlugin
from PIL.Image import Image as PILImageType
from send2trash import send2trash
from invokeai.app.models.image import ResourceOrigin
from invokeai.app.models.metadata import ImageMetadata
from invokeai.app.util.thumbnails import get_thumbnail_name, make_thumbnail
@ -60,7 +59,8 @@ class ImageFileStorageBase(ABC):
self,
image: PILImageType,
image_name: str,
metadata: Optional[ImageMetadata] = None,
metadata: Optional[dict] = None,
graph: Optional[dict] = None,
thumbnail_size: int = 256,
) -> None:
"""Saves an image and a 256x256 WEBP thumbnail. Returns a tuple of the image name, thumbnail name, and created timestamp."""
@ -80,7 +80,7 @@ class DiskImageFileStorage(ImageFileStorageBase):
__cache: Dict[Path, PILImageType]
__max_cache_size: int
def __init__(self, output_folder: str | Path):
def __init__(self, output_folder: Union[str, Path]):
self.__cache = dict()
self.__cache_ids = Queue()
self.__max_cache_size = 10 # TODO: get this from config
@ -111,20 +111,22 @@ class DiskImageFileStorage(ImageFileStorageBase):
self,
image: PILImageType,
image_name: str,
metadata: Optional[ImageMetadata] = None,
metadata: Optional[dict] = None,
graph: Optional[dict] = None,
thumbnail_size: int = 256,
) -> None:
try:
self.__validate_storage_folders()
image_path = self.get_path(image_name)
pnginfo = PngImagePlugin.PngInfo()
if metadata is not None:
pnginfo = PngImagePlugin.PngInfo()
pnginfo.add_text("invokeai", metadata.json())
image.save(image_path, "PNG", pnginfo=pnginfo)
else:
image.save(image_path, "PNG")
pnginfo.add_text("invokeai_metadata", json.dumps(metadata))
if graph is not None:
pnginfo.add_text("invokeai_graph", json.dumps(graph))
image.save(image_path, "PNG", pnginfo=pnginfo)
thumbnail_name = get_thumbnail_name(image_name)
thumbnail_path = self.get_path(thumbnail_name, thumbnail=True)
thumbnail_image = make_thumbnail(image, thumbnail_size)
@ -164,7 +166,7 @@ class DiskImageFileStorage(ImageFileStorageBase):
return path
def validate_path(self, path: str | Path) -> bool:
def validate_path(self, path: Union[str, Path]) -> bool:
"""Validates the path given for an image or thumbnail."""
path = path if isinstance(path, Path) else Path(path)
return path.exists()
@ -175,7 +177,7 @@ class DiskImageFileStorage(ImageFileStorageBase):
for folder in folders:
folder.mkdir(parents=True, exist_ok=True)
def __get_cache(self, image_name: Path) -> PILImageType | None:
def __get_cache(self, image_name: Path) -> Optional[PILImageType]:
return None if image_name not in self.__cache else self.__cache[image_name]
def __set_cache(self, image_name: Path, image: PILImageType):

91
invokeai/app/services/image_record_storage.py
View File

@ -1,23 +1,16 @@
import json
import sqlite3
import threading
from abc import ABC, abstractmethod
from datetime import datetime
from typing import Generic, Optional, TypeVar, cast
import sqlite3
import threading
from typing import Optional, Union
from pydantic import BaseModel, Field
from pydantic.generics import GenericModel
from invokeai.app.models.metadata import ImageMetadata
from invokeai.app.models.image import (
ImageCategory,
ResourceOrigin,
)
from invokeai.app.models.image import ImageCategory, ResourceOrigin
from invokeai.app.services.models.image_record import (
ImageRecord,
ImageRecordChanges,
deserialize_image_record,
)
ImageRecord, ImageRecordChanges, deserialize_image_record)
T = TypeVar("T", bound=BaseModel)
@ -55,6 +48,28 @@ class ImageRecordDeleteException(Exception):
super().__init__(message)
IMAGE_DTO_COLS = ", ".join(
list(
map(
lambda c: "images." + c,
[
"image_name",
"image_origin",
"image_category",
"width",
"height",
"session_id",
"node_id",
"is_intermediate",
"created_at",
"updated_at",
"deleted_at",
],
)
)
)
class ImageRecordStorageBase(ABC):
"""Low-level service responsible for interfacing with the image record store."""
@ -65,6 +80,11 @@ class ImageRecordStorageBase(ABC):
"""Gets an image record."""
pass
@abstractmethod
def get_metadata(self, image_name: str) -> Optional[dict]:
"""Gets an image's metadata'."""
pass
@abstractmethod
def update(
self,
@ -109,14 +129,14 @@ class ImageRecordStorageBase(ABC):
height: int,
session_id: Optional[str],
node_id: Optional[str],
metadata: Optional[ImageMetadata],
metadata: Optional[dict],
is_intermediate: bool = False,
) -> datetime:
"""Saves an image record."""
pass
@abstractmethod
def get_most_recent_image_for_board(self, board_id: str) -> ImageRecord | None:
def get_most_recent_image_for_board(self, board_id: str) -> Optional[ImageRecord]:
"""Gets the most recent image for a board."""
pass
@ -163,7 +183,6 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
node_id TEXT,
metadata TEXT,
is_intermediate BOOLEAN DEFAULT FALSE,
board_id TEXT,
created_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')),
-- Updated via trigger
updated_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')),
@ -208,19 +227,19 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
"""
)
def get(self, image_name: str) -> Union[ImageRecord, None]:
def get(self, image_name: str) -> Optional[ImageRecord]:
try:
self._lock.acquire()
self._cursor.execute(
f"""--sql
SELECT * FROM images
SELECT {IMAGE_DTO_COLS} FROM images
WHERE image_name = ?;
""",
(image_name,),
)
result = cast(Union[sqlite3.Row, None], self._cursor.fetchone())
result = cast(Optional[sqlite3.Row], self._cursor.fetchone())
except sqlite3.Error as e:
self._conn.rollback()
raise ImageRecordNotFoundException from e
@ -232,6 +251,28 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
return deserialize_image_record(dict(result))
def get_metadata(self, image_name: str) -> Optional[dict]:
try:
self._lock.acquire()
self._cursor.execute(
f"""--sql
SELECT images.metadata FROM images
WHERE image_name = ?;
""",
(image_name,),
)
result = cast(Optional[sqlite3.Row], self._cursor.fetchone())
if not result or not result[0]:
return None
return json.loads(result[0])
except sqlite3.Error as e:
self._conn.rollback()
raise ImageRecordNotFoundException from e
finally:
self._lock.release()
def update(
self,
image_name: str,
@ -299,8 +340,8 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
WHERE 1=1
"""
images_query = """--sql
SELECT images.*
images_query = f"""--sql
SELECT {IMAGE_DTO_COLS}
FROM images
LEFT JOIN board_images ON board_images.image_name = images.image_name
WHERE 1=1
@ -418,12 +459,12 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
width: int,
height: int,
node_id: Optional[str],
metadata: Optional[ImageMetadata],
metadata: Optional[dict],
is_intermediate: bool = False,
) -> datetime:
try:
metadata_json = (
None if metadata is None else metadata.json(exclude_none=True)
None if metadata is None else json.dumps(metadata)
)
self._lock.acquire()
self._cursor.execute(
@ -473,9 +514,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
finally:
self._lock.release()
def get_most_recent_image_for_board(
self, board_id: str
) -> Union[ImageRecord, None]:
def get_most_recent_image_for_board(self, board_id: str) -> Optional[ImageRecord]:
try:
self._lock.acquire()
self._cursor.execute(
@ -490,7 +529,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
(board_id,),
)
result = cast(Union[sqlite3.Row, None], self._cursor.fetchone())
result = cast(Optional[sqlite3.Row], self._cursor.fetchone())
finally:
self._lock.release()
if result is None:

116
invokeai/app/services/images.py
View File

@ -1,39 +1,30 @@
import json
from abc import ABC, abstractmethod
from logging import Logger
from typing import Optional, TYPE_CHECKING, Union
from typing import TYPE_CHECKING, Optional
from PIL.Image import Image as PILImageType
from invokeai.app.models.image import (
ImageCategory,
ResourceOrigin,
InvalidImageCategoryException,
InvalidOriginException,
)
from invokeai.app.models.metadata import ImageMetadata
from invokeai.app.services.board_image_record_storage import BoardImageRecordStorageBase
from invokeai.app.services.image_record_storage import (
ImageRecordDeleteException,
ImageRecordNotFoundException,
ImageRecordSaveException,
ImageRecordStorageBase,
OffsetPaginatedResults,
)
from invokeai.app.services.models.image_record import (
ImageRecord,
ImageDTO,
ImageRecordChanges,
image_record_to_dto,
)
from invokeai.app.invocations.metadata import ImageMetadata
from invokeai.app.models.image import (ImageCategory,
InvalidImageCategoryException,
InvalidOriginException, ResourceOrigin)
from invokeai.app.services.board_image_record_storage import \
BoardImageRecordStorageBase
from invokeai.app.services.graph import Graph
from invokeai.app.services.image_file_storage import (
ImageFileDeleteException,
ImageFileNotFoundException,
ImageFileSaveException,
ImageFileStorageBase,
)
from invokeai.app.services.item_storage import ItemStorageABC, PaginatedResults
from invokeai.app.services.metadata import MetadataServiceBase
ImageFileDeleteException, ImageFileNotFoundException,
ImageFileSaveException, ImageFileStorageBase)
from invokeai.app.services.image_record_storage import (
ImageRecordDeleteException, ImageRecordNotFoundException,
ImageRecordSaveException, ImageRecordStorageBase, OffsetPaginatedResults)
from invokeai.app.services.item_storage import ItemStorageABC
from invokeai.app.services.models.image_record import (ImageDTO, ImageRecord,
ImageRecordChanges,
image_record_to_dto)
from invokeai.app.services.resource_name import NameServiceBase
from invokeai.app.services.urls import UrlServiceBase
from invokeai.app.util.metadata import get_metadata_graph_from_raw_session
if TYPE_CHECKING:
from invokeai.app.services.graph import GraphExecutionState
@ -51,6 +42,7 @@ class ImageServiceABC(ABC):
node_id: Optional[str] = None,
session_id: Optional[str] = None,
is_intermediate: bool = False,
metadata: Optional[dict] = None,
) -> ImageDTO:
"""Creates an image, storing the file and its metadata."""
pass
@ -79,6 +71,11 @@ class ImageServiceABC(ABC):
"""Gets an image DTO."""
pass
@abstractmethod
def get_metadata(self, image_name: str) -> ImageMetadata:
"""Gets an image's metadata."""
pass
@abstractmethod
def get_path(self, image_name: str, thumbnail: bool = False) -> str:
"""Gets an image's path."""
@ -124,7 +121,6 @@ class ImageServiceDependencies:
image_records: ImageRecordStorageBase
image_files: ImageFileStorageBase
board_image_records: BoardImageRecordStorageBase
metadata: MetadataServiceBase
urls: UrlServiceBase
logger: Logger
names: NameServiceBase
@ -135,7 +131,6 @@ class ImageServiceDependencies:
image_record_storage: ImageRecordStorageBase,
image_file_storage: ImageFileStorageBase,
board_image_record_storage: BoardImageRecordStorageBase,
metadata: MetadataServiceBase,
url: UrlServiceBase,
logger: Logger,
names: NameServiceBase,
@ -144,7 +139,6 @@ class ImageServiceDependencies:
self.image_records = image_record_storage
self.image_files = image_file_storage
self.board_image_records = board_image_record_storage
self.metadata = metadata
self.urls = url
self.logger = logger
self.names = names
@ -165,6 +159,7 @@ class ImageService(ImageServiceABC):
node_id: Optional[str] = None,
session_id: Optional[str] = None,
is_intermediate: bool = False,
metadata: Optional[dict] = None,
) -> ImageDTO:
if image_origin not in ResourceOrigin:
raise InvalidOriginException
@ -174,7 +169,16 @@ class ImageService(ImageServiceABC):
image_name = self._services.names.create_image_name()
metadata = self._get_metadata(session_id, node_id)
graph = None
if session_id is not None:
session_raw = self._services.graph_execution_manager.get_raw(session_id)
if session_raw is not None:
try:
graph = get_metadata_graph_from_raw_session(session_raw)
except Exception as e:
self._services.logger.warn(f"Failed to parse session graph: {e}")
graph = None
(width, height) = image.size
@ -191,14 +195,12 @@ class ImageService(ImageServiceABC):
is_intermediate=is_intermediate,
# Nullable fields
node_id=node_id,
session_id=session_id,
metadata=metadata,
session_id=session_id,
)
self._services.image_files.save(
image_name=image_name,
image=image,
metadata=metadata,
image_name=image_name, image=image, metadata=metadata, graph=graph
)
image_dto = self.get_dto(image_name)
@ -268,6 +270,34 @@ class ImageService(ImageServiceABC):
self._services.logger.error("Problem getting image DTO")
raise e
def get_metadata(self, image_name: str) -> Optional[ImageMetadata]:
try:
image_record = self._services.image_records.get(image_name)
if not image_record.session_id:
return ImageMetadata()
session_raw = self._services.graph_execution_manager.get_raw(
image_record.session_id
)
graph = None
if session_raw:
try:
graph = get_metadata_graph_from_raw_session(session_raw)
except Exception as e:
self._services.logger.warn(f"Failed to parse session graph: {e}")
graph = None
metadata = self._services.image_records.get_metadata(image_name)
return ImageMetadata(graph=graph, metadata=metadata)
except ImageRecordNotFoundException:
self._services.logger.error("Image record not found")
raise
except Exception as e:
self._services.logger.error("Problem getting image DTO")
raise e
def get_path(self, image_name: str, thumbnail: bool = False) -> str:
try:
return self._services.image_files.get_path(image_name, thumbnail)
@ -367,15 +397,3 @@ class ImageService(ImageServiceABC):
except Exception as e:
self._services.logger.error("Problem deleting image records and files")
raise e
def _get_metadata(
self, session_id: Optional[str] = None, node_id: Optional[str] = None
) -> Union[ImageMetadata, None]:
"""Get the metadata for a node."""
metadata = None
if node_id is not None and session_id is not None:
session = self._services.graph_execution_manager.get(session_id)
metadata = self._services.metadata.create_image_metadata(session, node_id)
return metadata

6
invokeai/app/services/invocation_queue.py
View File

@ -5,7 +5,7 @@ from abc import ABC, abstractmethod
from queue import Queue
from pydantic import BaseModel, Field
from typing import Optional
class InvocationQueueItem(BaseModel):
graph_execution_state_id: str = Field(description="The ID of the graph execution state")
@ -22,7 +22,7 @@ class InvocationQueueABC(ABC):
pass
@abstractmethod
def put(self, item: InvocationQueueItem | None) -> None:
def put(self, item: Optional[InvocationQueueItem]) -> None:
pass
@abstractmethod
@ -57,7 +57,7 @@ class MemoryInvocationQueue(InvocationQueueABC):
return item
def put(self, item: InvocationQueueItem | None) -> None:
def put(self, item: Optional[InvocationQueueItem]) -> None:
self.__queue.put(item)
def cancel(self, graph_execution_state_id: str) -> None:

65
invokeai/app/services/invocation_services.py
View File

@ -7,13 +7,12 @@ if TYPE_CHECKING:
from invokeai.app.services.board_images import BoardImagesServiceABC
from invokeai.app.services.boards import BoardServiceABC
from invokeai.app.services.images import ImageServiceABC
from invokeai.backend import ModelManager
from invokeai.app.services.model_manager_service import ModelManagerServiceBase
from invokeai.app.services.events import EventServiceBase
from invokeai.app.services.latent_storage import LatentsStorageBase
from invokeai.app.services.restoration_services import RestorationServices
from invokeai.app.services.invocation_queue import InvocationQueueABC
from invokeai.app.services.item_storage import ItemStorageABC
from invokeai.app.services.config import InvokeAISettings
from invokeai.app.services.config import InvokeAIAppConfig
from invokeai.app.services.graph import GraphExecutionState, LibraryGraph
from invokeai.app.services.invoker import InvocationProcessorABC
@ -22,46 +21,44 @@ class InvocationServices:
"""Services that can be used by invocations"""
# TODO: Just forward-declared everything due to circular dependencies. Fix structure.
events: "EventServiceBase"
latents: "LatentsStorageBase"
queue: "InvocationQueueABC"
model_manager: "ModelManager"
restoration: "RestorationServices"
configuration: "InvokeAISettings"
images: "ImageServiceABC"
boards: "BoardServiceABC"
board_images: "BoardImagesServiceABC"
graph_library: "ItemStorageABC"["LibraryGraph"]
boards: "BoardServiceABC"
configuration: "InvokeAIAppConfig"
events: "EventServiceBase"
graph_execution_manager: "ItemStorageABC"["GraphExecutionState"]
graph_library: "ItemStorageABC"["LibraryGraph"]
images: "ImageServiceABC"
latents: "LatentsStorageBase"
logger: "Logger"
model_manager: "ModelManagerServiceBase"
processor: "InvocationProcessorABC"
queue: "InvocationQueueABC"
def __init__(
self,
model_manager: "ModelManager",
events: "EventServiceBase",
logger: "Logger",
latents: "LatentsStorageBase",
images: "ImageServiceABC",
boards: "BoardServiceABC",
board_images: "BoardImagesServiceABC",
queue: "InvocationQueueABC",
graph_library: "ItemStorageABC"["LibraryGraph"],
boards: "BoardServiceABC",
configuration: "InvokeAIAppConfig",
events: "EventServiceBase",
graph_execution_manager: "ItemStorageABC"["GraphExecutionState"],
graph_library: "ItemStorageABC"["LibraryGraph"],
images: "ImageServiceABC",
latents: "LatentsStorageBase",
logger: "Logger",
model_manager: "ModelManagerServiceBase",
processor: "InvocationProcessorABC",
restoration: "RestorationServices",
configuration: "InvokeAISettings",
queue: "InvocationQueueABC",
):
self.model_manager = model_manager
self.events = events
self.logger = logger
self.latents = latents
self.images = images
self.boards = boards
self.board_images = board_images
self.queue = queue
self.graph_library = graph_library
self.graph_execution_manager = graph_execution_manager
self.processor = processor
self.restoration = restoration
self.configuration = configuration
self.boards = boards
self.boards = boards
self.configuration = configuration
self.events = events
self.graph_execution_manager = graph_execution_manager
self.graph_library = graph_library
self.images = images
self.latents = latents
self.logger = logger
self.model_manager = model_manager
self.processor = processor
self.queue = queue

11
invokeai/app/services/invoker.py
View File

@ -1,14 +1,11 @@
# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
from abc import ABC
from threading import Event, Thread
from typing import Optional
from ..invocations.baseinvocation import InvocationContext
from .graph import Graph, GraphExecutionState
from .invocation_queue import InvocationQueueABC, InvocationQueueItem
from .invocation_queue import InvocationQueueItem
from .invocation_services import InvocationServices
from .item_storage import ItemStorageABC
class Invoker:
"""The invoker, used to execute invocations"""
@ -21,7 +18,7 @@ class Invoker:
def invoke(
self, graph_execution_state: GraphExecutionState, invoke_all: bool = False
) -> str | None:
) -> Optional[str]:
"""Determines the next node to invoke and enqueues it, preparing if needed.
Returns the id of the queued node, or `None` if there are no nodes left to enqueue."""
@ -45,7 +42,7 @@ class Invoker:
return invocation.id
def create_execution_state(self, graph: Graph | None = None) -> GraphExecutionState:
def create_execution_state(self, graph: Optional[Graph] = None) -> GraphExecutionState:
"""Creates a new execution state for the given graph"""
new_state = GraphExecutionState(graph=Graph() if graph is None else graph)
self.services.graph_execution_manager.set(new_state)

10
invokeai/app/services/item_storage.py
View File

@ -1,5 +1,5 @@
from abc import ABC, abstractmethod
from typing import Callable, Generic, TypeVar
from typing import Callable, Generic, Optional, TypeVar
from pydantic import BaseModel, Field
from pydantic.generics import GenericModel
@ -29,14 +29,22 @@ class ItemStorageABC(ABC, Generic[T]):
@abstractmethod
def get(self, item_id: str) -> T:
"""Gets the item, parsing it into a Pydantic model"""
pass
@abstractmethod
def get_raw(self, item_id: str) -> Optional[str]:
"""Gets the raw item as a string, skipping Pydantic parsing"""
pass
@abstractmethod
def set(self, item: T) -> None:
"""Sets the item"""
pass
@abstractmethod
def list(self, page: int = 0, per_page: int = 10) -> PaginatedResults[T]:
"""Gets a paginated list of items"""
pass
@abstractmethod

10
invokeai/app/services/latent_storage.py
View File

@ -3,7 +3,7 @@
from abc import ABC, abstractmethod
from pathlib import Path
from queue import Queue
from typing import Dict
from typing import Dict, Union, Optional
import torch
@ -55,7 +55,7 @@ class ForwardCacheLatentsStorage(LatentsStorageBase):
if name in self.__cache:
del self.__cache[name]
def __get_cache(self, name: str) -> torch.Tensor|None:
def __get_cache(self, name: str) -> Optional[torch.Tensor]:
return None if name not in self.__cache else self.__cache[name]
def __set_cache(self, name: str, data: torch.Tensor):
@ -69,9 +69,9 @@ class ForwardCacheLatentsStorage(LatentsStorageBase):
class DiskLatentsStorage(LatentsStorageBase):
"""Stores latents in a folder on disk without caching"""
__output_folder: str | Path
__output_folder: Union[str, Path]
def __init__(self, output_folder: str | Path):
def __init__(self, output_folder: Union[str, Path]):
self.__output_folder = output_folder if isinstance(output_folder, Path) else Path(output_folder)
self.__output_folder.mkdir(parents=True, exist_ok=True)
@ -91,4 +91,4 @@ class DiskLatentsStorage(LatentsStorageBase):
def get_path(self, name: str) -> Path:
return self.__output_folder / name

142
invokeai/app/services/metadata.py
View File

@ -1,142 +0,0 @@
from abc import ABC, abstractmethod
from typing import Any, Union
import networkx as nx
from invokeai.app.models.metadata import ImageMetadata
from invokeai.app.services.graph import Graph, GraphExecutionState
class MetadataServiceBase(ABC):
"""Handles building metadata for nodes, images, and outputs."""
@abstractmethod
def create_image_metadata(
self, session: GraphExecutionState, node_id: str
) -> ImageMetadata:
"""Builds an ImageMetadata object for a node."""
pass
class CoreMetadataService(MetadataServiceBase):
_ANCESTOR_TYPES = ["t2l", "l2l"]
"""The ancestor types that contain the core metadata"""
_ANCESTOR_PARAMS = ["type", "steps", "model", "cfg_scale", "scheduler", "strength"]
"""The core metadata parameters in the ancestor types"""
_NOISE_FIELDS = ["seed", "width", "height"]
"""The core metadata parameters in the noise node"""
def create_image_metadata(
self, session: GraphExecutionState, node_id: str
) -> ImageMetadata:
metadata = self._build_metadata_from_graph(session, node_id)
return metadata
def _find_nearest_ancestor(self, G: nx.DiGraph, node_id: str) -> Union[str, None]:
"""
Finds the id of the nearest ancestor (of a valid type) of a given node.
Parameters:
G (nx.DiGraph): The execution graph, converted in to a networkx DiGraph. Its nodes must
have the same data as the execution graph.
node_id (str): The ID of the node.
Returns:
str | None: The ID of the nearest ancestor, or None if there are no valid ancestors.
"""
# Retrieve the node from the graph
node = G.nodes[node_id]
# If the node type is one of the core metadata node types, return its id
if node.get("type") in self._ANCESTOR_TYPES:
return node.get("id")
# Else, look for the ancestor in the predecessor nodes
for predecessor in G.predecessors(node_id):
result = self._find_nearest_ancestor(G, predecessor)
if result:
return result
# If there are no valid ancestors, return None
return None
def _get_additional_metadata(
self, graph: Graph, node_id: str
) -> Union[dict[str, Any], None]:
"""
Returns additional metadata for a given node.
Parameters:
graph (Graph): The execution graph.
node_id (str): The ID of the node.
Returns:
dict[str, Any] | None: A dictionary of additional metadata.
"""
metadata = {}
# Iterate over all edges in the graph
for edge in graph.edges:
dest_node_id = edge.destination.node_id
dest_field = edge.destination.field
source_node_dict = graph.nodes[edge.source.node_id].dict()
# If the destination node ID matches the given node ID, gather necessary metadata
if dest_node_id == node_id:
# Prompt
if dest_field == "positive_conditioning":
metadata["positive_conditioning"] = source_node_dict.get("prompt")
# Negative prompt
if dest_field == "negative_conditioning":
metadata["negative_conditioning"] = source_node_dict.get("prompt")
# Seed, width and height
if dest_field == "noise":
for field in self._NOISE_FIELDS:
metadata[field] = source_node_dict.get(field)
return metadata
def _build_metadata_from_graph(
self, session: GraphExecutionState, node_id: str
) -> ImageMetadata:
"""
Builds an ImageMetadata object for a node.
Parameters:
session (GraphExecutionState): The session.
node_id (str): The ID of the node.
Returns:
ImageMetadata: The metadata for the node.
"""
# We need to do all the traversal on the execution graph
graph = session.execution_graph
# Find the nearest `t2l`/`l2l` ancestor of the given node
ancestor_id = self._find_nearest_ancestor(graph.nx_graph_with_data(), node_id)
# If no ancestor was found, return an empty ImageMetadata object
if ancestor_id is None:
return ImageMetadata()
ancestor_node = graph.get_node(ancestor_id)
# Grab all the core metadata from the ancestor node
ancestor_metadata = {
param: val
for param, val in ancestor_node.dict().items()
if param in self._ANCESTOR_PARAMS
}
# Get this image's prompts and noise parameters
addl_metadata = self._get_additional_metadata(graph, ancestor_id)
# If additional metadata was found, add it to the main metadata
if addl_metadata is not None:
ancestor_metadata.update(addl_metadata)
return ImageMetadata(**ancestor_metadata)

360
invokeai/app/services/model_manager_service.py
View File

@ -2,22 +2,30 @@
from __future__ import annotations
import torch
from abc import ABC, abstractmethod
from pathlib import Path
from typing import Optional, Union, Callable, List, Tuple, types, TYPE_CHECKING
from dataclasses import dataclass
from pydantic import Field
from typing import Optional, Union, Callable, List, Tuple, TYPE_CHECKING
from types import ModuleType
from invokeai.backend.model_management.model_manager import (
from invokeai.backend.model_management import (
ModelManager,
BaseModelType,
ModelType,
SubModelType,
ModelInfo,
AddModelResult,
SchedulerPredictionType,
ModelMerger,
MergeInterpolationMethod,
ModelNotFoundException,
)
from invokeai.backend.model_management.model_search import FindModels
import torch
from invokeai.app.models.exceptions import CanceledException
from .config import InvokeAIAppConfig
from ...backend.util import choose_precision, choose_torch_device
from .config import InvokeAIAppConfig
if TYPE_CHECKING:
from ..invocations.baseinvocation import BaseInvocation, InvocationContext
@ -30,16 +38,16 @@ class ModelManagerServiceBase(ABC):
def __init__(
self,
config: InvokeAIAppConfig,
logger: types.ModuleType,
logger: ModuleType,
):
"""
Initialize with the path to the models.yaml config file.
Initialize with the path to the models.yaml config file.
Optional parameters are the torch device type, precision, max_models,
and sequential_offload boolean. Note that the default device
type and precision are set up for a CUDA system running at half precision.
"""
pass
@abstractmethod
def get_model(
self,
@ -50,8 +58,8 @@ class ModelManagerServiceBase(ABC):
node: Optional[BaseInvocation] = None,
context: Optional[InvocationContext] = None,
) -> ModelInfo:
"""Retrieve the indicated model with name and type.
submodel can be used to get a part (such as the vae)
"""Retrieve the indicated model with name and type.
submodel can be used to get a part (such as the vae)
of a diffusers pipeline."""
pass
@ -73,13 +81,7 @@ class ModelManagerServiceBase(ABC):
def model_info(self, model_name: str, base_model: BaseModelType, model_type: ModelType) -> dict:
"""
Given a model name returns a dict-like (OmegaConf) object describing it.
"""
pass
@abstractmethod
def model_names(self) -> List[Tuple[str, BaseModelType, ModelType]]:
"""
Returns a list of all the model names known.
Uses the exact format as the omegaconf stanza.
"""
pass
@ -101,7 +103,20 @@ class ModelManagerServiceBase(ABC):
}
"""
pass
@abstractmethod
def list_model(self, model_name: str, base_model: BaseModelType, model_type: ModelType) -> dict:
"""
Return information about the model using the same format as list_models()
"""
pass
@abstractmethod
def model_names(self) -> List[Tuple[str, BaseModelType, ModelType]]:
"""
Returns a list of all the model names known.
"""
pass
@abstractmethod
def add_model(
@ -111,16 +126,34 @@ class ModelManagerServiceBase(ABC):
model_type: ModelType,
model_attributes: dict,
clobber: bool = False
) -> None:
) -> AddModelResult:
"""
Update the named model with a dictionary of attributes. Will fail with an
assertion error if the name already exists. Pass clobber=True to overwrite.
On a successful update, the config will be changed in memory. Will fail
with an assertion error if provided attributes are incorrect or
the model name is missing. Call commit() to write changes to disk.
"""
pass
@abstractmethod
def update_model(
self,
model_name: str,
base_model: BaseModelType,
model_type: ModelType,
model_attributes: dict,
) -> AddModelResult:
"""
Update the named model with a dictionary of attributes. Will fail with a
ModelNotFoundException if the name does not already exist.
On a successful update, the config will be changed in memory. Will fail
with an assertion error if provided attributes are incorrect or
the model name is missing. Call commit() to write changes to disk.
"""
pass
@abstractmethod
def del_model(
self,
@ -129,14 +162,117 @@ class ModelManagerServiceBase(ABC):
model_type: ModelType,
):
"""
Delete the named model from configuration. If delete_files is true,
then the underlying weight file or diffusers directory will be deleted
Delete the named model from configuration. If delete_files is true,
then the underlying weight file or diffusers directory will be deleted
as well. Call commit() to write to disk.
"""
pass
@abstractmethod
def commit(self, conf_file: Path = None) -> None:
def rename_model(self,
model_name: str,
base_model: BaseModelType,
model_type: ModelType,
new_name: str,
):
"""
Rename the indicated model.
"""
pass
@abstractmethod
def list_checkpoint_configs(
self
)->List[Path]:
"""
List the checkpoint config paths from ROOT/configs/stable-diffusion.
"""
pass
@abstractmethod
def convert_model(
self,
model_name: str,
base_model: BaseModelType,
model_type: Union[ModelType.Main,ModelType.Vae],
) -> AddModelResult:
"""
Convert a checkpoint file into a diffusers folder, deleting the cached
version and deleting the original checkpoint file if it is in the models
directory.
:param model_name: Name of the model to convert
:param base_model: Base model type
:param model_type: Type of model ['vae' or 'main']
This will raise a ValueError unless the model is not a checkpoint. It will
also raise a ValueError in the event that there is a similarly-named diffusers
directory already in place.
"""
pass
@abstractmethod
def heuristic_import(self,
items_to_import: set[str],
prediction_type_helper: Optional[Callable[[Path],SchedulerPredictionType]]=None,
)->dict[str, AddModelResult]:
'''Import a list of paths, repo_ids or URLs. Returns the set of
successfully imported items.
:param items_to_import: Set of strings corresponding to models to be imported.
:param prediction_type_helper: A callback that receives the Path of a Stable Diffusion 2 checkpoint model and returns a SchedulerPredictionType.
The prediction type helper is necessary to distinguish between
models based on Stable Diffusion 2 Base (requiring
SchedulerPredictionType.Epsilson) and Stable Diffusion 768
(requiring SchedulerPredictionType.VPrediction). It is
generally impossible to do this programmatically, so the
prediction_type_helper usually asks the user to choose.
The result is a set of successfully installed models. Each element
of the set is a dict corresponding to the newly-created OmegaConf stanza for
that model.
'''
pass
@abstractmethod
def merge_models(
self,
model_names: List[str] = Field(default=None, min_items=2, max_items=3, description="List of model names to merge"),
base_model: Union[BaseModelType,str] = Field(default=None, description="Base model shared by all models to be merged"),
merged_model_name: str = Field(default=None, description="Name of destination model after merging"),
alpha: Optional[float] = 0.5,
interp: Optional[MergeInterpolationMethod] = None,
force: Optional[bool] = False,
merge_dest_directory: Optional[Path] = None
) -> AddModelResult:
"""
Merge two to three diffusrs pipeline models and save as a new model.
:param model_names: List of 2-3 models to merge
:param base_model: Base model to use for all models
:param merged_model_name: Name of destination merged model
:param alpha: Alpha strength to apply to 2d and 3d model
:param interp: Interpolation method. None (default)
:param merge_dest_directory: Save the merged model to the designated directory (with 'merged_model_name' appended)
"""
pass
@abstractmethod
def search_for_models(self, directory: Path)->List[Path]:
"""
Return list of all models found in the designated directory.
"""
pass
@abstractmethod
def sync_to_config(self):
"""
Re-read models.yaml, rescan the models directory, and reimport models
in the autoimport directories. Call after making changes outside the
model manager API.
"""
pass
@abstractmethod
def commit(self, conf_file: Optional[Path] = None) -> None:
"""
Write current configuration out to the indicated file.
If no conf_file is provided, then replaces the
@ -150,10 +286,10 @@ class ModelManagerService(ModelManagerServiceBase):
def __init__(
self,
config: InvokeAIAppConfig,
logger: types.ModuleType,
logger: ModuleType,
):
"""
Initialize with the path to the models.yaml config file.
Initialize with the path to the models.yaml config file.
Optional parameters are the torch device type, precision, max_models,
and sequential_offload boolean. Note that the default device
type and precision are set up for a CUDA system running at half precision.
@ -162,12 +298,12 @@ class ModelManagerService(ModelManagerServiceBase):
config_file = config.model_conf_path
else:
config_file = config.root_dir / "configs/models.yaml"
if not config_file.exists():
raise IOError(f"The file {config_file} could not be found.")
logger.debug(f'config file={config_file}')
device = torch.device(choose_torch_device())
logger.debug(f'GPU device = {device}')
precision = config.precision
if precision == "auto":
precision = choose_precision(device)
@ -183,6 +319,8 @@ class ModelManagerService(ModelManagerServiceBase):
if hasattr(config,'max_cache_size') \
else config.max_loaded_models * 2.5
logger.debug(f"Maximum RAM cache size: {max_cache_size} GiB")
sequential_offload = config.sequential_guidance
self.mgr = ModelManager(
@ -238,7 +376,7 @@ class ModelManagerService(ModelManagerServiceBase):
submodel=submodel,
model_info=model_info
)
return model_info
def model_exists(
@ -274,12 +412,19 @@ class ModelManagerService(ModelManagerServiceBase):
base_model: Optional[BaseModelType] = None,
model_type: Optional[ModelType] = None
) -> list[dict]:
# ) -> dict:
"""
Return a list of models.
"""
return self.mgr.list_models(base_model, model_type)
def list_model(self, model_name: str, base_model: BaseModelType, model_type: ModelType) -> dict:
"""
Return information about the model using the same format as list_models()
"""
return self.mgr.list_model(model_name=model_name,
base_model=base_model,
model_type=model_type)
def add_model(
self,
model_name: str,
@ -291,13 +436,32 @@ class ModelManagerService(ModelManagerServiceBase):
"""
Update the named model with a dictionary of attributes. Will fail with an
assertion error if the name already exists. Pass clobber=True to overwrite.
On a successful update, the config will be changed in memory. Will fail
with an assertion error if provided attributes are incorrect or
the model name is missing. Call commit() to write changes to disk.
"""
self.logger.debug(f'add/update model {model_name}')
return self.mgr.add_model(model_name, base_model, model_type, model_attributes, clobber)
def update_model(
self,
model_name: str,
base_model: BaseModelType,
model_type: ModelType,
model_attributes: dict,
) -> AddModelResult:
"""
Update the named model with a dictionary of attributes. Will fail with a
ModelNotFoundException exception if the name does not already exist.
On a successful update, the config will be changed in memory. Will fail
with an assertion error if provided attributes are incorrect or
the model name is missing. Call commit() to write changes to disk.
"""
return self.mgr.add_model(model_name, base_model, model_type, model_attributes, clobber)
self.logger.debug(f'update model {model_name}')
if not self.model_exists(model_name, base_model, model_type):
raise ModelNotFoundException(f"Unknown model {model_name}")
return self.add_model(model_name, base_model, model_type, model_attributes, clobber=True)
def del_model(
self,
model_name: str,
@ -305,12 +469,36 @@ class ModelManagerService(ModelManagerServiceBase):
model_type: ModelType,
):
"""
Delete the named model from configuration. If delete_files is true,
then the underlying weight file or diffusers directory will be deleted
as well. Call commit() to write to disk.
Delete the named model from configuration. If delete_files is true,
then the underlying weight file or diffusers directory will be deleted
as well.
"""
self.logger.debug(f'delete model {model_name}')
self.mgr.del_model(model_name, base_model, model_type)
self.mgr.commit()
def convert_model(
self,
model_name: str,
base_model: BaseModelType,
model_type: Union[ModelType.Main,ModelType.Vae],
convert_dest_directory: Optional[Path] = Field(default=None, description="Optional directory location for merged model"),
) -> AddModelResult:
"""
Convert a checkpoint file into a diffusers folder, deleting the cached
version and deleting the original checkpoint file if it is in the models
directory.
:param model_name: Name of the model to convert
:param base_model: Base model type
:param model_type: Type of model ['vae' or 'main']
:param convert_dest_directory: Save the converted model to the designated directory (`models/etc/etc` by default)
This will raise a ValueError unless the model is not a checkpoint. It will
also raise a ValueError in the event that there is a similarly-named diffusers
directory already in place.
"""
self.logger.debug(f'convert model {model_name}')
return self.mgr.convert_model(model_name, base_model, model_type, convert_dest_directory)
def commit(self, conf_file: Optional[Path]=None):
"""
@ -360,4 +548,106 @@ class ModelManagerService(ModelManagerServiceBase):
@property
def logger(self):
return self.mgr.logger
def heuristic_import(self,
items_to_import: set[str],
prediction_type_helper: Optional[Callable[[Path],SchedulerPredictionType]]=None,
)->dict[str, AddModelResult]:
'''Import a list of paths, repo_ids or URLs. Returns the set of
successfully imported items.
:param items_to_import: Set of strings corresponding to models to be imported.
:param prediction_type_helper: A callback that receives the Path of a Stable Diffusion 2 checkpoint model and returns a SchedulerPredictionType.
The prediction type helper is necessary to distinguish between
models based on Stable Diffusion 2 Base (requiring
SchedulerPredictionType.Epsilson) and Stable Diffusion 768
(requiring SchedulerPredictionType.VPrediction). It is
generally impossible to do this programmatically, so the
prediction_type_helper usually asks the user to choose.
The result is a set of successfully installed models. Each element
of the set is a dict corresponding to the newly-created OmegaConf stanza for
that model.
'''
return self.mgr.heuristic_import(items_to_import, prediction_type_helper)
def merge_models(
self,
model_names: List[str] = Field(default=None, min_items=2, max_items=3, description="List of model names to merge"),
base_model: Union[BaseModelType,str] = Field(default=None, description="Base model shared by all models to be merged"),
merged_model_name: str = Field(default=None, description="Name of destination model after merging"),
alpha: Optional[float] = 0.5,
interp: Optional[MergeInterpolationMethod] = None,
force: Optional[bool] = False,
merge_dest_directory: Optional[Path] = Field(default=None, description="Optional directory location for merged model"),
) -> AddModelResult:
"""
Merge two to three diffusrs pipeline models and save as a new model.
:param model_names: List of 2-3 models to merge
:param base_model: Base model to use for all models
:param merged_model_name: Name of destination merged model
:param alpha: Alpha strength to apply to 2d and 3d model
:param interp: Interpolation method. None (default)
:param merge_dest_directory: Save the merged model to the designated directory (with 'merged_model_name' appended)
"""
merger = ModelMerger(self.mgr)
try:
result = merger.merge_diffusion_models_and_save(
model_names = model_names,
base_model = base_model,
merged_model_name = merged_model_name,
alpha = alpha,
interp = interp,
force = force,
merge_dest_directory=merge_dest_directory,
)
except AssertionError as e:
raise ValueError(e)
return result
def search_for_models(self, directory: Path)->List[Path]:
"""
Return list of all models found in the designated directory.
"""
search = FindModels(directory,self.logger)
return search.list_models()
def sync_to_config(self):
"""
Re-read models.yaml, rescan the models directory, and reimport models
in the autoimport directories. Call after making changes outside the
model manager API.
"""
return self.mgr.sync_to_config()
def list_checkpoint_configs(self)->List[Path]:
"""
List the checkpoint config paths from ROOT/configs/stable-diffusion.
"""
config = self.mgr.app_config
conf_path = config.legacy_conf_path
root_path = config.root_path
return [(conf_path / x).relative_to(root_path) for x in conf_path.glob('**/*.yaml')]
def rename_model(self,
model_name: str,
base_model: BaseModelType,
model_type: ModelType,
new_name: str = None,
new_base: BaseModelType = None,
):
"""
Rename the indicated model. Can provide a new name and/or a new base.
:param model_name: Current name of the model
:param base_model: Current base of the model
:param model_type: Model type (can't be changed)
:param new_name: New name for the model
:param new_base: New base for the model
"""
self.mgr.rename_model(base_model = base_model,
model_type = model_type,
model_name = model_name,
new_name = new_name,
new_base = new_base,
)

23
invokeai/app/services/models/image_record.py
View File

@ -1,13 +1,14 @@
import datetime
from typing import Optional, Union
from pydantic import BaseModel, Extra, Field, StrictBool, StrictStr
from invokeai.app.models.image import ImageCategory, ResourceOrigin
from invokeai.app.models.metadata import ImageMetadata
from invokeai.app.util.misc import get_iso_timestamp
class ImageRecord(BaseModel):
"""Deserialized image record."""
"""Deserialized image record without metadata."""
image_name: str = Field(description="The unique name of the image.")
"""The unique name of the image."""
@ -43,11 +44,6 @@ class ImageRecord(BaseModel):
description="The node ID that generated this image, if it is a generated image.",
)
"""The node ID that generated this image, if it is a generated image."""
metadata: Optional[ImageMetadata] = Field(
default=None,
description="A limited subset of the image's generation metadata. Retrieve the image's session for full metadata.",
)
"""A limited subset of the image's generation metadata. Retrieve the image's session for full metadata."""
class ImageRecordChanges(BaseModel, extra=Extra.forbid):
@ -88,7 +84,7 @@ class ImageUrlsDTO(BaseModel):
class ImageDTO(ImageRecord, ImageUrlsDTO):
"""Deserialized image record, enriched for the frontend."""
board_id: Union[str, None] = Field(
board_id: Optional[str] = Field(
description="The id of the board the image belongs to, if one exists."
)
"""The id of the board the image belongs to, if one exists."""
@ -96,7 +92,7 @@ class ImageDTO(ImageRecord, ImageUrlsDTO):
def image_record_to_dto(
image_record: ImageRecord, image_url: str, thumbnail_url: str, board_id: Union[str, None]
image_record: ImageRecord, image_url: str, thumbnail_url: str, board_id: Optional[str]
) -> ImageDTO:
"""Converts an image record to an image DTO."""
return ImageDTO(
@ -112,6 +108,7 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
# Retrieve all the values, setting "reasonable" defaults if they are not present.
# TODO: do we really need to handle default values here? ideally the data is the correct shape...
image_name = image_dict.get("image_name", "unknown")
image_origin = ResourceOrigin(
image_dict.get("image_origin", ResourceOrigin.INTERNAL.value)
@ -128,13 +125,6 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
deleted_at = image_dict.get("deleted_at", get_iso_timestamp())
is_intermediate = image_dict.get("is_intermediate", False)
raw_metadata = image_dict.get("metadata")
if raw_metadata is not None:
metadata = ImageMetadata.parse_raw(raw_metadata)
else:
metadata = None
return ImageRecord(
image_name=image_name,
image_origin=image_origin,
@ -143,7 +133,6 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
height=height,
session_id=session_id,
node_id=node_id,
metadata=metadata,
created_at=created_at,
updated_at=updated_at,
deleted_at=deleted_at,

1
invokeai/app/services/processor.py
View File

@ -104,6 +104,7 @@ class DefaultInvocationProcessor(InvocationProcessorABC):
except Exception as e:
error = traceback.format_exc()
logger.error(error)
# Save error
graph_execution_state.set_node_error(invocation.id, error)

113
invokeai/app/services/restoration_services.py
View File

@ -1,113 +0,0 @@
import sys
import traceback
import torch
from typing import types
from ...backend.restoration import Restoration
from ...backend.util import choose_torch_device, CPU_DEVICE, MPS_DEVICE
# This should be a real base class for postprocessing functions,
# but right now we just instantiate the existing gfpgan, esrgan
# and codeformer functions.
class RestorationServices:
'''Face restoration and upscaling'''
def __init__(self,args,logger:types.ModuleType):
try:
gfpgan, codeformer, esrgan = None, None, None
if args.restore or args.esrgan:
restoration = Restoration()
# TODO: redo for new model structure
if False and args.restore:
gfpgan, codeformer = restoration.load_face_restore_models(
args.gfpgan_model_path
)
else:
logger.info("Face restoration disabled")
if False and args.esrgan:
esrgan = restoration.load_esrgan(args.esrgan_bg_tile)
else:
logger.info("Upscaling disabled")
else:
logger.info("Face restoration and upscaling disabled")
except (ModuleNotFoundError, ImportError):
print(traceback.format_exc(), file=sys.stderr)
logger.info("You may need to install the ESRGAN and/or GFPGAN modules")
self.device = torch.device(choose_torch_device())
self.gfpgan = gfpgan
self.codeformer = codeformer
self.esrgan = esrgan
self.logger = logger
self.logger.info('Face restoration initialized')
# note that this one method does gfpgan and codepath reconstruction, as well as
# esrgan upscaling
# TO DO: refactor into separate methods
def upscale_and_reconstruct(
self,
image_list,
facetool="gfpgan",
upscale=None,
upscale_denoise_str=0.75,
strength=0.0,
codeformer_fidelity=0.75,
save_original=False,
image_callback=None,
prefix=None,
):
results = []
for r in image_list:
image, seed = r
try:
if strength > 0:
if self.gfpgan is not None or self.codeformer is not None:
if facetool == "gfpgan":
if self.gfpgan is None:
self.logger.info(
"GFPGAN not found. Face restoration is disabled."
)
else:
image = self.gfpgan.process(image, strength, seed)
if facetool == "codeformer":
if self.codeformer is None:
self.logger.info(
"CodeFormer not found. Face restoration is disabled."
)
else:
cf_device = (
CPU_DEVICE if self.device == MPS_DEVICE else self.device
)
image = self.codeformer.process(
image=image,
strength=strength,
device=cf_device,
seed=seed,
fidelity=codeformer_fidelity,
)
else:
self.logger.info("Face Restoration is disabled.")
if upscale is not None:
if self.esrgan is not None:
if len(upscale) < 2:
upscale.append(0.75)
image = self.esrgan.process(
image,
upscale[1],
seed,
int(upscale[0]),
denoise_str=upscale_denoise_str,
)
else:
self.logger.info("ESRGAN is disabled. Image not upscaled.")
except Exception as e:
self.logger.info(
f"Error running RealESRGAN or GFPGAN. Your image was not upscaled.\n{e}"
)
if image_callback is not None:
image_callback(image, seed, upscaled=True, use_prefix=prefix)
else:
r[0] = image
results.append([image, seed])
return results

19
invokeai/app/services/sqlite.py
View File

@ -1,6 +1,6 @@
import sqlite3
from threading import Lock
from typing import Generic, TypeVar, Union, get_args
from typing import Generic, Optional, TypeVar, get_args
from pydantic import BaseModel, parse_raw_as
@ -63,7 +63,7 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
self._lock.release()
self._on_changed(item)
def get(self, id: str) -> Union[T, None]:
def get(self, id: str) -> Optional[T]:
try:
self._lock.acquire()
self._cursor.execute(
@ -78,6 +78,21 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
return self._parse_item(result[0])
def get_raw(self, id: str) -> Optional[str]:
try:
self._lock.acquire()
self._cursor.execute(
f"""SELECT item FROM {self._table_name} WHERE id = ?;""", (str(id),)
)
result = self._cursor.fetchone()
finally:
self._lock.release()
if not result:
return None
return result[0]
def delete(self, id: str):
try:
self._lock.acquire()

2
invokeai/app/services/urls.py
View File

@ -22,4 +22,4 @@ class LocalUrlService(UrlServiceBase):
if thumbnail:
return f"{self._base_url}/images/{image_basename}/thumbnail"
return f"{self._base_url}/images/{image_basename}"
return f"{self._base_url}/images/{image_basename}/full"

55
invokeai/app/util/metadata.py Normal file
View File

@ -0,0 +1,55 @@
import json
from typing import Optional
from pydantic import ValidationError
from invokeai.app.services.graph import Edge
def get_metadata_graph_from_raw_session(session_raw: str) -> Optional[dict]:
"""
Parses raw session string, returning a dict of the graph.
Only the general graph shape is validated; none of the fields are validated.
Any `metadata_accumulator` nodes and edges are removed.
Any validation failure will return None.
"""
graph = json.loads(session_raw).get("graph", None)
# sanity check make sure the graph is at least reasonably shaped
if (
type(graph) is not dict
or "nodes" not in graph
or type(graph["nodes"]) is not dict
or "edges" not in graph
or type(graph["edges"]) is not list
):
# something has gone terribly awry, return an empty dict
return None
try:
# delete the `metadata_accumulator` node
del graph["nodes"]["metadata_accumulator"]
except KeyError:
# no accumulator node, all good
pass
# delete any edges to or from it
for i, edge in enumerate(graph["edges"]):
try:
# try to parse the edge
Edge(**edge)
except ValidationError:
# something has gone terribly awry, return an empty dict
return None
if (
edge["source"]["node_id"] == "metadata_accumulator"
or edge["destination"]["node_id"] == "metadata_accumulator"
):
del graph["edges"][i]
return graph

38
invokeai/backend/generator/base.py
View File

@ -21,7 +21,7 @@ from PIL import Image, ImageChops, ImageFilter
from accelerate.utils import set_seed
from diffusers import DiffusionPipeline
from tqdm import trange
from typing import Callable, List, Iterator, Optional, Type
from typing import Callable, List, Iterator, Optional, Type, Union
from dataclasses import dataclass, field
from diffusers.schedulers import SchedulerMixin as Scheduler
@ -178,7 +178,7 @@ class InvokeAIGenerator(metaclass=ABCMeta):
# ------------------------------------
class Img2Img(InvokeAIGenerator):
def generate(self,
init_image: Image.Image | torch.FloatTensor,
init_image: Union[Image.Image, torch.FloatTensor],
strength: float=0.75,
**keyword_args
)->Iterator[InvokeAIGeneratorOutput]:
@ -195,7 +195,7 @@ class Img2Img(InvokeAIGenerator):
# Takes all the arguments of Img2Img and adds the mask image and the seam/infill stuff
class Inpaint(Img2Img):
def generate(self,
mask_image: Image.Image | torch.FloatTensor,
mask_image: Union[Image.Image, torch.FloatTensor],
# Seam settings - when 0, doesn't fill seam
seam_size: int = 96,
seam_blur: int = 16,
@ -570,28 +570,16 @@ class Generator:
device = self.model.device
# limit noise to only the diffusion image channels, not the mask channels
input_channels = min(self.latent_channels, 4)
if self.use_mps_noise or device.type == "mps":
x = torch.randn(
[
1,
input_channels,
height // self.downsampling_factor,
width // self.downsampling_factor,
],
dtype=self.torch_dtype(),
device="cpu",
).to(device)
else:
x = torch.randn(
[
1,
input_channels,
height // self.downsampling_factor,
width // self.downsampling_factor,
],
dtype=self.torch_dtype(),
device=device,
)
x = torch.randn(
[
1,
input_channels,
height // self.downsampling_factor,
width // self.downsampling_factor,
],
dtype=self.torch_dtype(),
device=device,
)
if self.perlin > 0.0:
perlin_noise = self.get_perlin_noise(
width // self.downsampling_factor, height // self.downsampling_factor

5
invokeai/backend/generator/img2img.py
View File

@ -88,10 +88,7 @@ class Img2Img(Generator):
def get_noise_like(self, like: torch.Tensor):
device = like.device
if device.type == "mps":
x = torch.randn_like(like, device="cpu").to(device)
else:
x = torch.randn_like(like, device=device)
x = torch.randn_like(like, device=device)
if self.perlin > 0.0:
shape = like.shape
x = (1 - self.perlin) * x + self.perlin * self.get_perlin_noise(

9
invokeai/backend/generator/inpaint.py
View File

@ -4,11 +4,10 @@ invokeai.backend.generator.inpaint descends from .generator
from __future__ import annotations
import math
from typing import Tuple, Union
from typing import Tuple, Union, Optional
import cv2
import numpy as np
import PIL
import torch
from PIL import Image, ImageChops, ImageFilter, ImageOps
@ -76,7 +75,7 @@ class Inpaint(Img2Img):
return im_patched
def tile_fill_missing(
self, im: Image.Image, tile_size: int = 16, seed: Union[int, None] = None
self, im: Image.Image, tile_size: int = 16, seed: Optional[int] = None
) -> Image.Image:
# Only fill if there's an alpha layer
if im.mode != "RGBA":
@ -203,8 +202,8 @@ class Inpaint(Img2Img):
cfg_scale,
ddim_eta,
conditioning,
init_image: Image.Image | torch.FloatTensor,
mask_image: Image.Image | torch.FloatTensor,
init_image: Union[Image.Image, torch.FloatTensor],
mask_image: Union[Image.Image, torch.FloatTensor],
strength: float,
mask_blur_radius: int = 8,
# Seam settings - when 0, doesn't fill seam

104
invokeai/backend/install/invokeai_configure.py
View File

@ -30,8 +30,6 @@ from huggingface_hub import login as hf_hub_login
from omegaconf import OmegaConf
from tqdm import tqdm
from transformers import (
AutoProcessor,
CLIPSegForImageSegmentation,
CLIPTextModel,
CLIPTokenizer,
AutoFeatureExtractor,
@ -76,7 +74,7 @@ Weights_dir = "ldm/stable-diffusion-v1/"
Default_config_file = config.model_conf_path
SD_Configs = config.legacy_conf_path
PRECISION_CHOICES = ['auto','float16','float32','autocast']
PRECISION_CHOICES = ['auto','float16','float32']
INIT_FILE_PREAMBLE = """# InvokeAI initialization file
# This is the InvokeAI initialization file, which contains command-line default values.
@ -225,64 +223,30 @@ def download_conversion_models():
# ---------------------------------------------
def download_realesrgan():
logger.info("Installing models from RealESRGAN...")
model_url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth"
wdn_model_url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth"
model_dest = config.root_path / "models/core/upscaling/realesrgan/realesr-general-x4v3.pth"
wdn_model_dest = config.root_path / "models/core/upscaling/realesrgan/realesr-general-wdn-x4v3.pth"
download_with_progress_bar(model_url, str(model_dest), "RealESRGAN")
download_with_progress_bar(wdn_model_url, str(wdn_model_dest), "RealESRGANwdn")
def download_gfpgan():
logger.info("Installing GFPGAN models...")
for model in (
[
"https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth",
"./models/core/face_restoration/gfpgan/GFPGANv1.4.pth",
],
[
"https://github.com/xinntao/facexlib/releases/download/v0.1.0/detection_Resnet50_Final.pth",
"./models/core/face_restoration/gfpgan/weights/detection_Resnet50_Final.pth",
],
[
"https://github.com/xinntao/facexlib/releases/download/v0.2.2/parsing_parsenet.pth",
"./models/core/face_restoration/gfpgan/weights/parsing_parsenet.pth",
],
):
model_url, model_dest = model[0], config.root_path / model[1]
download_with_progress_bar(model_url, str(model_dest), "GFPGAN weights")
logger.info("Installing RealESRGAN models...")
URLs = [
dict(
url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth",
dest = "core/upscaling/realesrgan/RealESRGAN_x4plus.pth",
description = "RealESRGAN_x4plus.pth",
),
dict(
url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth",
dest = "core/upscaling/realesrgan/RealESRGAN_x4plus_anime_6B.pth",
description = "RealESRGAN_x4plus_anime_6B.pth",
),
dict(
url= "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
dest= "core/upscaling/realesrgan/ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
description = "ESRGAN_SRx4_DF2KOST_official.pth",
),
]
for model in URLs:
download_with_progress_bar(model['url'], config.models_path / model['dest'], model['description'])
# ---------------------------------------------
def download_codeformer():
logger.info("Installing CodeFormer model file...")
model_url = (
"https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/codeformer.pth"
)
model_dest = config.root_path / "models/core/face_restoration/codeformer/codeformer.pth"
download_with_progress_bar(model_url, str(model_dest), "CodeFormer")
# ---------------------------------------------
def download_clipseg():
logger.info("Installing clipseg model for text-based masking...")
CLIPSEG_MODEL = "CIDAS/clipseg-rd64-refined"
try:
hf_download_from_pretrained(AutoProcessor, CLIPSEG_MODEL, config.root_path / 'models/core/misc/clipseg')
hf_download_from_pretrained(CLIPSegForImageSegmentation, CLIPSEG_MODEL, config.root_path / 'models/core/misc/clipseg')
except Exception:
logger.info("Error installing clipseg model:")
logger.info(traceback.format_exc())
def download_support_models():
download_realesrgan()
download_gfpgan()
download_codeformer()
download_clipseg()
download_conversion_models()
# -------------------------------------
@ -359,9 +323,7 @@ Use cursor arrows to make a checkbox selection, and space to toggle.
scroll_exit=True,
)
self.nextrely += 1
label = """If you have an account at HuggingFace you may optionally paste your access token here
to allow InvokeAI to download restricted styles & subjects from the "Concept Library". See https://huggingface.co/settings/tokens.
"""
label = """HuggingFace access token (OPTIONAL) for automatic model downloads. See https://huggingface.co/settings/tokens."""
for line in textwrap.wrap(label,width=window_width-6):
self.add_widget_intelligent(
npyscreen.FixedText,
@ -423,6 +385,7 @@ to allow InvokeAI to download restricted styles & subjects from the "Concept Lib
)
self.precision = self.add_widget_intelligent(
npyscreen.TitleSelectOne,
columns = 2,
name="Precision",
values=PRECISION_CHOICES,
value=PRECISION_CHOICES.index(precision),
@ -430,13 +393,13 @@ to allow InvokeAI to download restricted styles & subjects from the "Concept Lib
max_height=len(PRECISION_CHOICES) + 1,
scroll_exit=True,
)
self.max_loaded_models = self.add_widget_intelligent(
self.max_cache_size = self.add_widget_intelligent(
IntTitleSlider,
name="Number of models to cache in CPU memory (each will use 2-4 GB!)",
value=old_opts.max_loaded_models,
out_of=10,
lowest=1,
begin_entry_at=4,
name="Size of the RAM cache used for fast model switching (GB)",
value=old_opts.max_cache_size,
out_of=20,
lowest=3,
begin_entry_at=6,
scroll_exit=True,
)
self.nextrely += 1
@ -539,7 +502,7 @@ https://huggingface.co/spaces/CompVis/stable-diffusion-license
"outdir",
"nsfw_checker",
"free_gpu_mem",
"max_loaded_models",
"max_cache_size",
"xformers_enabled",
"always_use_cpu",
]:
@ -555,9 +518,6 @@ https://huggingface.co/spaces/CompVis/stable-diffusion-license
new_opts.license_acceptance = self.license_acceptance.value
new_opts.precision = PRECISION_CHOICES[self.precision.value[0]]
# widget library workaround to make max_loaded_models an int rather than a float
new_opts.max_loaded_models = int(new_opts.max_loaded_models)
return new_opts
@ -861,9 +821,9 @@ def main():
download_support_models()
if opt.skip_sd_weights:
logger.info("\n** SKIPPING DIFFUSION WEIGHTS DOWNLOAD PER USER REQUEST **")
logger.warning("SKIPPING DIFFUSION WEIGHTS DOWNLOAD PER USER REQUEST")
elif models_to_download:
logger.info("\n** DOWNLOADING DIFFUSION WEIGHTS **")
logger.info("DOWNLOADING DIFFUSION WEIGHTS")
process_and_execute(opt, models_to_download)
postscript(errors=errors)

11
invokeai/backend/install/legacy_arg_parsing.py
View File

@ -4,6 +4,8 @@ import argparse
import shlex
from argparse import ArgumentParser
# note that this includes both old sampler names and new scheduler names
# in order to be able to parse both 2.0 and 3.0-pre-nodes versions of invokeai.init
SAMPLER_CHOICES = [
"ddim",
"ddpm",
@ -27,6 +29,15 @@ SAMPLER_CHOICES = [
"dpmpp_sde",
"dpmpp_sde_k",
"unipc",
"k_dpm_2_a",
"k_dpm_2",
"k_dpmpp_2_a",
"k_dpmpp_2",
"k_euler_a",
"k_euler",
"k_heun",
"k_lms",
"plms",
]
PRECISION_CHOICES = [

292
invokeai/backend/install/migrate_to_3.py
View File

@ -3,7 +3,6 @@ Migrate the models directory and models.yaml file from an existing
InvokeAI 2.3 installation to 3.0.0.
'''
import io
import os
import argparse
import shutil
@ -28,9 +27,10 @@ from transformers import (
)
import invokeai.backend.util.logging as logger
from invokeai.app.services.config import InvokeAIAppConfig
from invokeai.backend.model_management import ModelManager
from invokeai.backend.model_management.model_probe import (
ModelProbe, ModelType, BaseModelType, SchedulerPredictionType, ModelProbeInfo
ModelProbe, ModelType, BaseModelType, ModelProbeInfo
)
warnings.filterwarnings("ignore")
@ -47,48 +47,27 @@ class ModelPaths:
class MigrateTo3(object):
def __init__(self,
root_directory: Path,
dest_models: Path,
yaml_file: io.TextIOBase,
from_root: Path,
to_models: Path,
model_manager: ModelManager,
src_paths: ModelPaths,
):
self.root_directory = root_directory
self.dest_models = dest_models
self.dest_yaml = yaml_file
self.model_names = set()
self.root_directory = from_root
self.dest_models = to_models
self.mgr = model_manager
self.src_paths = src_paths
self._initialize_yaml()
def _initialize_yaml(self):
self.dest_yaml.write(
yaml.dump(
{
'__metadata__':
@classmethod
def initialize_yaml(cls, yaml_file: Path):
with open(yaml_file, 'w') as file:
file.write(
yaml.dump(
{
'version':'3.0.0'}
}
'__metadata__': {'version':'3.0.0'}
}
)
)
)
def unique_name(self,name,info)->str:
'''
Create a unique name for a model for use within models.yaml.
'''
done = False
key = ModelManager.create_key(name,info.base_type,info.model_type)
unique_name = key
counter = 1
while not done:
if unique_name in self.model_names:
unique_name = f'{key}-{counter:0>2d}'
counter += 1
else:
done = True
self.model_names.add(unique_name)
name,_,_ = ModelManager.parse_key(unique_name)
return name
def create_directory_structure(self):
'''
Create the basic directory structure for the models folder.
@ -136,23 +115,8 @@ class MigrateTo3(object):
that looks like a model, and copy the model into the
appropriate location within the destination models directory.
'''
directories_scanned = set()
for root, dirs, files in os.walk(src_dir):
for f in files:
# hack - don't copy raw learned_embeds.bin, let them
# be copied as part of a tree copy operation
if f == 'learned_embeds.bin':
continue
try:
model = Path(root,f)
info = ModelProbe().heuristic_probe(model)
if not info:
continue
dest = self._model_probe_to_path(info) / f
self.copy_file(model, dest)
except KeyboardInterrupt:
raise
except Exception as e:
logger.error(str(e))
for d in dirs:
try:
model = Path(root,d)
@ -161,6 +125,29 @@ class MigrateTo3(object):
continue
dest = self._model_probe_to_path(info) / model.name
self.copy_dir(model, dest)
directories_scanned.add(model)
except Exception as e:
logger.error(str(e))
except KeyboardInterrupt:
raise
except Exception as e:
logger.error(str(e))
for f in files:
# don't copy raw learned_embeds.bin or pytorch_lora_weights.bin
# let them be copied as part of a tree copy operation
try:
if f in {'learned_embeds.bin','pytorch_lora_weights.bin'}:
continue
model = Path(root,f)
if model.parent in directories_scanned:
continue
info = ModelProbe().heuristic_probe(model)
if not info:
continue
dest = self._model_probe_to_path(info) / f
self.copy_file(model, dest)
except Exception as e:
logger.error(str(e))
except KeyboardInterrupt:
raise
except Exception as e:
@ -219,11 +206,12 @@ class MigrateTo3(object):
repo_id = 'openai/clip-vit-large-patch14'
self._migrate_pretrained(CLIPTokenizer,
repo_id= repo_id,
dest= target_dir / 'clip-vit-large-patch14' / 'tokenizer',
dest= target_dir / 'clip-vit-large-patch14',
**kwargs)
self._migrate_pretrained(CLIPTextModel,
repo_id = repo_id,
dest = target_dir / 'clip-vit-large-patch14' / 'text_encoder',
dest = target_dir / 'clip-vit-large-patch14',
force = True,
**kwargs)
# sd-2
@ -262,46 +250,24 @@ class MigrateTo3(object):
except Exception as e:
logger.error(str(e))
def write_yaml(self, model_name: str, path:Path, info:ModelProbeInfo, **kwargs):
'''
Write a stanza for a moved model into the new models.yaml file.
'''
name = self.unique_name(model_name, info)
stanza = {
f'{info.base_type.value}/{info.model_type.value}/{name}': {
'name': model_name,
'path': str(path),
'description': f'A {info.base_type.value} {info.model_type.value} model',
'format': info.format,
'image_size': info.image_size,
'base': info.base_type.value,
'variant': info.variant_type.value,
'prediction_type': info.prediction_type.value,
'upcast_attention': info.prediction_type == SchedulerPredictionType.VPrediction,
**kwargs,
}
}
self.dest_yaml.write(yaml.dump(stanza))
self.dest_yaml.flush()
def _model_probe_to_path(self, info: ModelProbeInfo)->Path:
return Path(self.dest_models, info.base_type.value, info.model_type.value)
def _migrate_pretrained(self, model_class, repo_id: str, dest: Path, **kwargs):
if dest.exists():
def _migrate_pretrained(self, model_class, repo_id: str, dest: Path, force:bool=False, **kwargs):
if dest.exists() and not force:
logger.info(f'Skipping existing {dest}')
return
model = model_class.from_pretrained(repo_id, **kwargs)
self._save_pretrained(model, dest)
self._save_pretrained(model, dest, overwrite=force)
def _save_pretrained(self, model, dest: Path):
if dest.exists():
logger.info(f'Skipping existing {dest}')
return
def _save_pretrained(self, model, dest: Path, overwrite: bool=False):
model_name = dest.name
download_path = dest.with_name(f'{model_name}.downloading')
model.save_pretrained(download_path, safe_serialization=True)
download_path.replace(dest)
if overwrite:
model.save_pretrained(dest, safe_serialization=True)
else:
download_path = dest.with_name(f'{model_name}.downloading')
model.save_pretrained(download_path, safe_serialization=True)
download_path.replace(dest)
def _download_vae(self, repo_id: str, subfolder:str=None)->Path:
vae = AutoencoderKL.from_pretrained(repo_id, cache_dir=self.root_directory / 'models/hub', subfolder=subfolder)
@ -327,6 +293,7 @@ class MigrateTo3(object):
elif repo_id := vae.get('repo_id'):
if repo_id=='stabilityai/sd-vae-ft-mse': # this guy is already downloaded
vae_path = 'models/core/convert/sd-vae-ft-mse'
return vae_path
else:
vae_path = self._download_vae(repo_id, vae.get('subfolder'))
@ -339,7 +306,10 @@ class MigrateTo3(object):
info = ModelProbe().heuristic_probe(vae_path)
dest = self._model_probe_to_path(info) / vae_path.name
if not dest.exists():
self.copy_dir(vae_path,dest)
if vae_path.is_dir():
self.copy_dir(vae_path,dest)
else:
self.copy_file(vae_path,dest)
vae_path = dest
if vae_path.is_relative_to(self.dest_models):
@ -348,7 +318,7 @@ class MigrateTo3(object):
else:
return vae_path
def migrate_repo_id(self, repo_id: str, model_name :str=None, **extra_config):
def migrate_repo_id(self, repo_id: str, model_name: str=None, **extra_config):
'''
Migrate a locally-cached diffusers pipeline identified with a repo_id
'''
@ -380,11 +350,15 @@ class MigrateTo3(object):
if not info:
return
dest = self._model_probe_to_path(info) / repo_name
if self.mgr.model_exists(model_name, info.base_type, info.model_type):
logger.warning(f'A model named {model_name} already exists at the destination. Skipping migration.')
return
dest = self._model_probe_to_path(info) / model_name
self._save_pretrained(pipeline, dest)
rel_path = Path('models',dest.relative_to(dest_dir))
self.write_yaml(model_name, path=rel_path, info=info, **extra_config)
self._add_model(model_name, info, rel_path, **extra_config)
def migrate_path(self, location: Path, model_name: str=None, **extra_config):
'''
@ -394,20 +368,49 @@ class MigrateTo3(object):
# handle relative paths
dest_dir = self.dest_models
location = self.root_directory / location
model_name = model_name or location.stem
info = ModelProbe().heuristic_probe(location)
if not info:
return
if self.mgr.model_exists(model_name, info.base_type, info.model_type):
logger.warning(f'A model named {model_name} already exists at the destination. Skipping migration.')
return
# uh oh, weights is in the old models directory - move it into the new one
if Path(location).is_relative_to(self.src_paths.models):
dest = Path(dest_dir, info.base_type.value, info.model_type.value, location.name)
self.copy_dir(location,dest)
if location.is_dir():
self.copy_dir(location,dest)
else:
self.copy_file(location,dest)
location = Path('models', info.base_type.value, info.model_type.value, location.name)
model_name = model_name or location.stem
model_name = self.unique_name(model_name, info)
self.write_yaml(model_name, path=location, info=info, **extra_config)
self._add_model(model_name, info, location, **extra_config)
def _add_model(self,
model_name: str,
info: ModelProbeInfo,
location: Path,
**extra_config):
if info.model_type != ModelType.Main:
return
self.mgr.add_model(
model_name = model_name,
base_model = info.base_type,
model_type = info.model_type,
clobber = True,
model_attributes = {
'path': str(location),
'description': f'A {info.base_type.value} {info.model_type.value} model',
'model_format': info.format,
'variant': info.variant_type.value,
**extra_config,
}
)
def migrate_defined_models(self):
'''
Migrate models defined in models.yaml
@ -429,6 +432,9 @@ class MigrateTo3(object):
if config := stanza.get('config'):
passthru_args['config'] = config
if description:= stanza.get('description'):
passthru_args['description'] = description
if repo_id := stanza.get('repo_id'):
logger.info(f'Migrating diffusers model {model_name}')
@ -509,31 +515,50 @@ def get_legacy_embeddings(root: Path) -> ModelPaths:
return _parse_legacy_yamlfile(root, path)
def do_migrate(src_directory: Path, dest_directory: Path):
"""
Migrate models from src to dest InvokeAI root directories
"""
config_file = dest_directory / 'configs' / 'models.yaml.3'
dest_models = dest_directory / 'models.3'
dest_models = dest_directory / 'models-3.0'
dest_yaml = dest_directory / 'configs/models.yaml-3.0'
version_3 = (dest_directory / 'models' / 'core').exists()
# Here we create the destination models.yaml file.
# If we are writing into a version 3 directory and the
# file already exists, then we write into a copy of it to
# avoid deleting its previous customizations. Otherwise we
# create a new empty one.
if version_3: # write into the dest directory
try:
shutil.copy(dest_directory / 'configs' / 'models.yaml', config_file)
except:
MigrateTo3.initialize_yaml(config_file)
mgr = ModelManager(config_file) # important to initialize BEFORE moving the models directory
(dest_directory / 'models').replace(dest_models)
else:
MigrateTo3.initialize_yaml(config_file)
mgr = ModelManager(config_file)
paths = get_legacy_embeddings(src_directory)
migrator = MigrateTo3(
from_root = src_directory,
to_models = dest_models,
model_manager = mgr,
src_paths = paths
)
migrator.migrate()
print("Migration successful.")
with open(dest_yaml,'w') as yaml_file:
migrator = MigrateTo3(src_directory,
dest_models,
yaml_file,
src_paths = paths,
)
migrator.migrate()
shutil.rmtree(dest_directory / 'models.orig', ignore_errors=True)
(dest_directory / 'models').replace(dest_directory / 'models.orig')
dest_models.replace(dest_directory / 'models')
(dest_directory /'configs/models.yaml').replace(dest_directory / 'configs/models.yaml.orig')
dest_yaml.replace(dest_directory / 'configs/models.yaml')
print(f"""Migration successful.
Original models directory moved to {dest_directory}/models.orig
Original models.yaml file moved to {dest_directory}/configs/models.yaml.orig
""")
if not version_3:
(dest_directory / 'models').replace(src_directory / 'models.orig')
print(f'Original models directory moved to {dest_directory}/models.orig')
(dest_directory / 'configs' / 'models.yaml').replace(src_directory / 'configs' / 'models.yaml.orig')
print(f'Original models.yaml file moved to {dest_directory}/configs/models.yaml.orig')
config_file.replace(config_file.with_suffix(''))
dest_models.replace(dest_models.with_suffix(''))
def main():
parser = argparse.ArgumentParser(prog="invokeai-migrate3",
description="""
@ -545,34 +570,37 @@ It is safe to provide the same directory for both arguments, but it is better to
script, which will perform a full upgrade in place."""
)
parser.add_argument('--from-directory',
dest='root_directory',
dest='src_root',
type=Path,
required=True,
help='Source InvokeAI 2.3 root directory (containing "invokeai.init" or "invokeai.yaml")'
)
parser.add_argument('--to-directory',
dest='dest_directory',
dest='dest_root',
type=Path,
required=True,
help='Destination InvokeAI 3.0 directory (containing "invokeai.yaml")'
)
# TO DO: Implement full directory scanning
# parser.add_argument('--all-models',
# action="store_true",
# help='Migrate all models found in `models` directory, not just those mentioned in models.yaml',
# )
args = parser.parse_args()
root_directory = args.root_directory
assert root_directory.is_dir(), f"{root_directory} is not a valid directory"
assert (root_directory / 'models').is_dir(), f"{root_directory} does not contain a 'models' subdirectory"
assert (root_directory / 'invokeai.init').exists() or (root_directory / 'invokeai.yaml').exists(), f"{root_directory} does not contain an InvokeAI init file."
src_root = args.src_root
assert src_root.is_dir(), f"{src_root} is not a valid directory"
assert (src_root / 'models').is_dir(), f"{src_root} does not contain a 'models' subdirectory"
assert (src_root / 'models' / 'hub').exists(), f"{src_root} does not contain a version 2.3 models directory"
assert (src_root / 'invokeai.init').exists() or (src_root / 'invokeai.yaml').exists(), f"{src_root} does not contain an InvokeAI init file."
dest_directory = args.dest_directory
assert dest_directory.is_dir(), f"{dest_directory} is not a valid directory"
assert (dest_directory / 'models').is_dir(), f"{dest_directory} does not contain a 'models' subdirectory"
assert (dest_directory / 'invokeai.yaml').exists(), f"{dest_directory} does not contain an InvokeAI init file."
dest_root = args.dest_root
assert dest_root.is_dir(), f"{dest_root} is not a valid directory"
config = InvokeAIAppConfig.get_config()
config.parse_args(['--root',str(dest_root)])
do_migrate(root_directory,dest_directory)
# TODO: revisit - don't rely on invokeai.yaml to exist yet!
dest_is_setup = (dest_root / 'models/core').exists() and (dest_root / 'databases').exists()
if not dest_is_setup:
import invokeai.frontend.install.invokeai_configure
from invokeai.backend.install.invokeai_configure import initialize_rootdir
initialize_rootdir(dest_root, True)
do_migrate(src_root,dest_root)
if __name__ == '__main__':
main()

121
invokeai/backend/install/model_install_backend.py
View File

@ -19,7 +19,7 @@ from tqdm import tqdm
import invokeai.configs as configs
from invokeai.app.services.config import InvokeAIAppConfig
from invokeai.backend.model_management import ModelManager, ModelType, BaseModelType, ModelVariantType
from invokeai.backend.model_management import ModelManager, ModelType, BaseModelType, ModelVariantType, AddModelResult
from invokeai.backend.model_management.model_probe import ModelProbe, SchedulerPredictionType, ModelProbeInfo
from invokeai.backend.util import download_with_resume
from ..util.logging import InvokeAILogger
@ -71,8 +71,6 @@ class ModelInstallList:
class InstallSelections():
install_models: List[str]= field(default_factory=list)
remove_models: List[str]=field(default_factory=list)
# scan_directory: Path = None
# autoscan_on_startup: bool=False
@dataclass
class ModelLoadInfo():
@ -119,10 +117,11 @@ class ModelInstall(object):
# supplement with entries in models.yaml
installed_models = self.mgr.list_models()
for md in installed_models:
base = md['base_model']
model_type = md['type']
name = md['name']
model_type = md['model_type']
name = md['model_name']
key = ModelManager.create_key(name, base, model_type)
if key in model_dict:
model_dict[key].installed = True
@ -136,6 +135,12 @@ class ModelInstall(object):
)
return {x : model_dict[x] for x in sorted(model_dict.keys(),key=lambda y: model_dict[y].name.lower())}
def list_models(self, model_type):
installed = self.mgr.list_models(model_type=model_type)
print(f'Installed models of type `{model_type}`:')
for i in installed:
print(f"{i['model_name']}\t{i['base_model']}\t{i['path']}")
def starter_models(self)->Set[str]:
models = set()
for key, value in self.datasets.items():
@ -173,74 +178,78 @@ class ModelInstall(object):
# add requested models
for path in selections.install_models:
logger.info(f'Installing {path} [{job}/{jobs}]')
self.heuristic_install(path)
try:
self.heuristic_import(path)
except (ValueError, KeyError) as e:
logger.error(str(e))
job += 1
dlogging.set_verbosity(verbosity)
self.mgr.commit()
def heuristic_install(self,
model_path_id_or_url: Union[str,Path],
models_installed: Set[Path]=None)->Set[Path]:
def heuristic_import(self,
model_path_id_or_url: Union[str,Path],
models_installed: Set[Path]=None,
)->Dict[str, AddModelResult]:
'''
:param model_path_id_or_url: A Path to a local model to import, or a string representing its repo_id or URL
:param models_installed: Set of installed models, used for recursive invocation
Returns a set of dict objects corresponding to newly-created stanzas in models.yaml.
'''
if not models_installed:
models_installed = set()
models_installed = dict()
# A little hack to allow nested routines to retrieve info on the requested ID
self.current_id = model_path_id_or_url
path = Path(model_path_id_or_url)
# checkpoint file, or similar
if path.is_file():
models_installed.update({str(path):self._install_path(path)})
try:
# checkpoint file, or similar
if path.is_file():
models_installed.add(self._install_path(path))
# folders style or similar
elif path.is_dir() and any([(path/x).exists() for x in \
{'config.json','model_index.json','learned_embeds.bin','pytorch_lora_weights.bin'}
]
):
models_installed.update(self._install_path(path))
# folders style or similar
elif path.is_dir() and any([(path/x).exists() for x in {'config.json','model_index.json','learned_embeds.bin'}]):
models_installed.add(self._install_path(path))
# recursive scan
elif path.is_dir():
for child in path.iterdir():
self.heuristic_import(child, models_installed=models_installed)
# recursive scan
elif path.is_dir():
for child in path.iterdir():
self.heuristic_install(child, models_installed=models_installed)
# huggingface repo
elif len(str(model_path_id_or_url).split('/')) == 2:
models_installed.update({str(model_path_id_or_url): self._install_repo(str(model_path_id_or_url))})
# huggingface repo
elif len(str(path).split('/')) == 2:
models_installed.add(self._install_repo(str(path)))
# a URL
elif str(model_path_id_or_url).startswith(("http:", "https:", "ftp:")):
models_installed.update({str(model_path_id_or_url): self._install_url(model_path_id_or_url)})
# a URL
elif model_path_id_or_url.startswith(("http:", "https:", "ftp:")):
models_installed.add(self._install_url(model_path_id_or_url))
else:
logger.warning(f'{str(model_path_id_or_url)} is not recognized as a local path, repo ID or URL. Skipping')
except ValueError as e:
logger.error(str(e))
else:
raise KeyError(f'{str(model_path_id_or_url)} is not recognized as a local path, repo ID or URL. Skipping')
return models_installed
# install a model from a local path. The optional info parameter is there to prevent
# the model from being probed twice in the event that it has already been probed.
def _install_path(self, path: Path, info: ModelProbeInfo=None)->Path:
try:
# logger.debug(f'Probing {path}')
info = info or ModelProbe().heuristic_probe(path,self.prediction_helper)
model_name = path.stem if info.format=='checkpoint' else path.name
if self.mgr.model_exists(model_name, info.base_type, info.model_type):
raise ValueError(f'A model named "{model_name}" is already installed.')
attributes = self._make_attributes(path,info)
self.mgr.add_model(model_name = model_name,
base_model = info.base_type,
model_type = info.model_type,
model_attributes = attributes,
)
except Exception as e:
logger.warning(f'{str(e)} Skipping registration.')
return path
def _install_path(self, path: Path, info: ModelProbeInfo=None)->AddModelResult:
info = info or ModelProbe().heuristic_probe(path,self.prediction_helper)
if not info:
logger.warning(f'Unable to parse format of {path}')
return None
model_name = path.stem if path.is_file() else path.name
if self.mgr.model_exists(model_name, info.base_type, info.model_type):
raise ValueError(f'A model named "{model_name}" is already installed.')
attributes = self._make_attributes(path,info)
return self.mgr.add_model(model_name = model_name,
base_model = info.base_type,
model_type = info.model_type,
model_attributes = attributes,
)
def _install_url(self, url: str)->Path:
# copy to a staging area, probe, import and delete
def _install_url(self, url: str)->AddModelResult:
with TemporaryDirectory(dir=self.config.models_path) as staging:
location = download_with_resume(url,Path(staging))
if not location:
@ -252,7 +261,7 @@ class ModelInstall(object):
# staged version will be garbage-collected at this time
return self._install_path(Path(models_path), info)
def _install_repo(self, repo_id: str)->Path:
def _install_repo(self, repo_id: str)->AddModelResult:
hinfo = HfApi().model_info(repo_id)
# we try to figure out how to download this most economically
@ -278,16 +287,16 @@ class ModelInstall(object):
location = self._download_hf_model(repo_id, files, staging)
break
elif f'learned_embeds.{suffix}' in files:
location = self._download_hf_model(repo_id, ['learned_embeds.suffix'], staging)
location = self._download_hf_model(repo_id, [f'learned_embeds.{suffix}'], staging)
break
if not location:
logger.warning(f'Could not determine type of repo {repo_id}. Skipping install.')
return
return {}
info = ModelProbe().heuristic_probe(location, self.prediction_helper)
if not info:
logger.warning(f'Could not probe {location}. Skipping install.')
return
return {}
dest = self.config.models_path / info.base_type.value / info.model_type.value / self._get_model_name(repo_id,location)
if dest.exists():
shutil.rmtree(dest)

5
invokeai/backend/model_management/__init__.py
View File

@ -1,7 +1,8 @@
"""
Initialization file for invokeai.backend.model_management
"""
from .model_manager import ModelManager, ModelInfo
from .model_manager import ModelManager, ModelInfo, AddModelResult, SchedulerPredictionType
from .model_cache import ModelCache
from .models import BaseModelType, ModelType, SubModelType, ModelVariantType
from .models import BaseModelType, ModelType, SubModelType, ModelVariantType, ModelNotFoundException
from .model_merge import ModelMerger, MergeInterpolationMethod

7
invokeai/backend/model_management/convert_ckpt_to_diffusers.py
View File

@ -29,7 +29,7 @@ import invokeai.backend.util.logging as logger
from invokeai.app.services.config import InvokeAIAppConfig
from .model_manager import ModelManager
from .model_cache import ModelCache
from picklescan.scanner import scan_file_path
from .models import BaseModelType, ModelVariantType
try:
@ -1014,7 +1014,10 @@ def load_pipeline_from_original_stable_diffusion_ckpt(
checkpoint = load_file(checkpoint_path)
else:
if scan_needed:
ModelCache.scan_model(checkpoint_path, checkpoint_path)
# scan model
scan_result = scan_file_path(checkpoint_path)
if scan_result.infected_files != 0:
raise "The model {checkpoint_path} is potentially infected by malware. Aborting import."
checkpoint = torch.load(checkpoint_path)
# sometimes there is a state_dict key and sometimes not

47
invokeai/backend/model_management/lora.py
View File

@ -1,18 +1,15 @@
from __future__ import annotations
import copy
from pathlib import Path
from contextlib import contextmanager
from typing import Optional, Dict, Tuple, Any
from typing import Optional, Dict, Tuple, Any, Union, List
from pathlib import Path
import torch
from safetensors.torch import load_file
from torch.utils.hooks import RemovableHandle
from diffusers.models import UNet2DConditionModel
from transformers import CLIPTextModel
from compel.embeddings_provider import BaseTextualInversionManager
from diffusers.models import UNet2DConditionModel
from safetensors.torch import load_file
from transformers import CLIPTextModel, CLIPTokenizer
class LoRALayerBase:
#rank: Optional[int]
@ -124,8 +121,8 @@ class LoRALayer(LoRALayerBase):
def get_weight(self):
if self.mid is not None:
up = self.up.reshape(up.shape[0], up.shape[1])
down = self.down.reshape(up.shape[0], up.shape[1])
up = self.up.reshape(self.up.shape[0], self.up.shape[1])
down = self.down.reshape(self.down.shape[0], self.down.shape[1])
weight = torch.einsum("m n w h, i m, n j -> i j w h", self.mid, up, down)
else:
weight = self.up.reshape(self.up.shape[0], -1) @ self.down.reshape(self.down.shape[0], -1)
@ -411,7 +408,7 @@ class LoRAModel: #(torch.nn.Module):
else:
# TODO: diff/ia3/... format
print(
f">> Encountered unknown lora layer module in {self.name}: {layer_key}"
f">> Encountered unknown lora layer module in {model.name}: {layer_key}"
)
return
@ -539,9 +536,10 @@ class ModelPatcher:
original_weights[module_key] = module.weight.detach().to(device="cpu", copy=True)
# enable autocast to calc fp16 loras on cpu
with torch.autocast(device_type="cpu"):
layer_scale = layer.alpha / layer.rank if (layer.alpha and layer.rank) else 1.0
layer_weight = layer.get_weight() * lora_weight * layer_scale
#with torch.autocast(device_type="cpu"):
layer.to(dtype=torch.float32)
layer_scale = layer.alpha / layer.rank if (layer.alpha and layer.rank) else 1.0
layer_weight = layer.get_weight() * lora_weight * layer_scale
if module.weight.shape != layer_weight.shape:
# TODO: debug on lycoris
@ -617,6 +615,24 @@ class ModelPatcher:
text_encoder.resize_token_embeddings(init_tokens_count)
@classmethod
@contextmanager
def apply_clip_skip(
cls,
text_encoder: CLIPTextModel,
clip_skip: int,
):
skipped_layers = []
try:
for i in range(clip_skip):
skipped_layers.append(text_encoder.text_model.encoder.layers.pop(-1))
yield
finally:
while len(skipped_layers) > 0:
text_encoder.text_model.encoder.layers.append(skipped_layers.pop())
class TextualInversionModel:
name: str
embedding: torch.Tensor # [n, 768]|[n, 1280]
@ -655,6 +671,9 @@ class TextualInversionModel:
else:
result.embedding = next(iter(state_dict.values()))
if len(result.embedding.shape) == 1:
result.embedding = result.embedding.unsqueeze(0)
if not isinstance(result.embedding, torch.Tensor):
raise ValueError(f"Invalid embeddings file: {file_path.name}")

55
invokeai/backend/model_management/model_cache.py
View File

@ -8,7 +8,7 @@ The cache returns context manager generators designed to load the
model into the GPU within the context, and unload outside the
context. Use like this:
cache = ModelCache(max_models_cached=6)
cache = ModelCache(max_cache_size=7.5)
with cache.get_model('runwayml/stable-diffusion-1-5') as SD1,
cache.get_model('stabilityai/stable-diffusion-2') as SD2:
do_something_in_GPU(SD1,SD2)
@ -36,6 +36,9 @@ from .models import BaseModelType, ModelType, SubModelType, ModelBase
# Default is roughly enough to hold three fp16 diffusers models in RAM simultaneously
DEFAULT_MAX_CACHE_SIZE = 6.0
# amount of GPU memory to hold in reserve for use by generations (GB)
DEFAULT_MAX_VRAM_CACHE_SIZE= 2.75
# actual size of a gig
GIG = 1073741824
@ -82,6 +85,7 @@ class ModelCache(object):
def __init__(
self,
max_cache_size: float=DEFAULT_MAX_CACHE_SIZE,
max_vram_cache_size: float=DEFAULT_MAX_VRAM_CACHE_SIZE,
execution_device: torch.device=torch.device('cuda'),
storage_device: torch.device=torch.device('cpu'),
precision: torch.dtype=torch.float16,
@ -91,7 +95,7 @@ class ModelCache(object):
logger: types.ModuleType = logger
):
'''
:param max_models: Maximum number of models to cache in CPU RAM [4]
:param max_cache_size: Maximum size of the RAM cache [6.0 GB]
:param execution_device: Torch device to load active model into [torch.device('cuda')]
:param storage_device: Torch device to save inactive model in [torch.device('cpu')]
:param precision: Precision for loaded models [torch.float16]
@ -99,14 +103,11 @@ class ModelCache(object):
:param sequential_offload: Conserve VRAM by loading and unloading each stage of the pipeline sequentially
:param sha_chunksize: Chunksize to use when calculating sha256 model hash
'''
#max_cache_size = 9999
execution_device = torch.device('cuda')
self.model_infos: Dict[str, ModelBase] = dict()
self.lazy_offloading = lazy_offloading
#self.sequential_offload: bool=sequential_offload
self.precision: torch.dtype=precision
self.max_cache_size: int=max_cache_size
self.max_cache_size: float=max_cache_size
self.max_vram_cache_size: float=max_vram_cache_size
self.execution_device: torch.device=execution_device
self.storage_device: torch.device=storage_device
self.sha_chunksize=sha_chunksize
@ -128,16 +129,6 @@ class ModelCache(object):
key += f":{submodel_type}"
return key
#def get_model(
# self,
# repo_id_or_path: Union[str, Path],
# model_type: ModelType = ModelType.Diffusers,
# subfolder: Path = None,
# submodel: ModelType = None,
# revision: str = None,
# attach_model_part: Tuple[ModelType, str] = (None, None),
# gpu_load: bool = True,
#) -> ModelLocker: # ?? what does it return
def _get_model_info(
self,
model_path: str,
@ -213,14 +204,22 @@ class ModelCache(object):
self._cache_stack.remove(key)
self._cache_stack.append(key)
return self.ModelLocker(self, key, cache_entry.model, gpu_load)
return self.ModelLocker(self, key, cache_entry.model, gpu_load, cache_entry.size)
class ModelLocker(object):
def __init__(self, cache, key, model, gpu_load):
def __init__(self, cache, key, model, gpu_load, size_needed):
'''
:param cache: The model_cache object
:param key: The key of the model to lock in GPU
:param model: The model to lock
:param gpu_load: True if load into gpu
:param size_needed: Size of the model to load
'''
self.gpu_load = gpu_load
self.cache = cache
self.key = key
self.model = model
self.size_needed = size_needed
self.cache_entry = self.cache._cached_models[self.key]
def __enter__(self) -> Any:
@ -234,7 +233,7 @@ class ModelCache(object):
try:
if self.cache.lazy_offloading:
self.cache._offload_unlocked_models()
self.cache._offload_unlocked_models(self.size_needed)
if self.model.device != self.cache.execution_device:
self.cache.logger.debug(f'Moving {self.key} into {self.cache.execution_device}')
@ -349,12 +348,20 @@ class ModelCache(object):
self.logger.debug(f"After unloading: cached_models={len(self._cached_models)}")
def _offload_unlocked_models(self):
for model_key, cache_entry in self._cached_models.items():
def _offload_unlocked_models(self, size_needed: int=0):
reserved = self.max_vram_cache_size * GIG
vram_in_use = torch.cuda.memory_allocated()
self.logger.debug(f'{(vram_in_use/GIG):.2f}GB VRAM used for models; max allowed={(reserved/GIG):.2f}GB')
for model_key, cache_entry in sorted(self._cached_models.items(), key=lambda x:x[1].size):
if vram_in_use <= reserved:
break
if not cache_entry.locked and cache_entry.loaded:
self.logger.debug(f'Offloading {model_key} from {self.execution_device} into {self.storage_device}')
cache_entry.model.to(self.storage_device)
with VRAMUsage() as mem:
cache_entry.model.to(self.storage_device)
self.logger.debug(f'GPU VRAM freed: {(mem.vram_used/GIG):.2f} GB')
vram_in_use += mem.vram_used # note vram_used is negative
self.logger.debug(f'{(vram_in_use/GIG):.2f}GB VRAM used for models; max allowed={(reserved/GIG):.2f}GB')
def _local_model_hash(self, model_path: Union[str, Path]) -> str:
sha = hashlib.sha256()

344
invokeai/backend/model_management/model_manager.py
View File

@ -231,26 +231,28 @@ from __future__ import annotations
import os
import hashlib
import textwrap
import yaml
from dataclasses import dataclass
from pathlib import Path
from typing import Optional, List, Tuple, Union, Set, Callable, types
from shutil import rmtree
from typing import Optional, List, Tuple, Union, Dict, Set, Callable, types
from shutil import rmtree, move
import torch
from omegaconf import OmegaConf
from omegaconf.dictconfig import DictConfig
from pydantic import BaseModel
from pydantic import BaseModel, Field
import invokeai.backend.util.logging as logger
from invokeai.app.services.config import InvokeAIAppConfig
from invokeai.backend.util import CUDA_DEVICE, Chdir
from .model_cache import ModelCache, ModelLocker
from .model_search import ModelSearch
from .models import (
BaseModelType, ModelType, SubModelType,
ModelError, SchedulerPredictionType, MODEL_CLASSES,
ModelConfigBase,
)
ModelConfigBase, ModelNotFoundException, InvalidModelException,
)
# We are only starting to number the config file with release 3.
# The config file version doesn't have to start at release version, but it will help
@ -274,13 +276,14 @@ class ModelInfo():
def __exit__(self,*args, **kwargs):
self.context.__exit__(*args, **kwargs)
class InvalidModelError(Exception):
"Raised when an invalid model is requested"
pass
class AddModelResult(BaseModel):
name: str = Field(description="The name of the model after installation")
model_type: ModelType = Field(description="The type of model")
base_model: BaseModelType = Field(description="The base model")
config: ModelConfigBase = Field(description="The configuration of the model")
MAX_CACHE_SIZE = 6.0 # GB
class ConfigMeta(BaseModel):
version: str
@ -306,10 +309,12 @@ class ModelManager(object):
and sequential_offload boolean. Note that the default device
type and precision are set up for a CUDA system running at half precision.
"""
self.config_path = None
if isinstance(config, (str, Path)):
self.config_path = Path(config)
if not self.config_path.exists():
logger.warning(f'The file {self.config_path} was not found. Initializing a new file')
self.initialize_model_config(self.config_path)
config = OmegaConf.load(self.config_path)
elif not isinstance(config, DictConfig):
@ -318,9 +323,31 @@ class ModelManager(object):
self.config_meta = ConfigMeta(**config.pop("__metadata__"))
# TODO: metadata not found
# TODO: version check
self.app_config = InvokeAIAppConfig.get_config()
self.logger = logger
self.cache = ModelCache(
max_cache_size=max_cache_size,
max_vram_cache_size = self.app_config.max_vram_cache_size,
execution_device = device_type,
precision = precision,
sequential_offload = sequential_offload,
logger = logger,
)
self._read_models(config)
def _read_models(self, config: Optional[DictConfig] = None):
if not config:
if self.config_path:
config = OmegaConf.load(self.config_path)
else:
return
self.models = dict()
for model_key, model_config in config.items():
if model_key.startswith('_'):
continue
model_name, base_model, model_type = self.parse_key(model_key)
model_class = MODEL_CLASSES[base_model][model_type]
# alias for config file
@ -328,20 +355,20 @@ class ModelManager(object):
self.models[model_key] = model_class.create_config(**model_config)
# check config version number and update on disk/RAM if necessary
self.app_config = InvokeAIAppConfig.get_config()
self.logger = logger
self.cache = ModelCache(
max_cache_size=max_cache_size,
execution_device = device_type,
precision = precision,
sequential_offload = sequential_offload,
logger = logger,
)
self.cache_keys = dict()
# add controlnet, lora and textual_inversion models from disk
self.scan_models_directory()
def sync_to_config(self):
"""
Call this when `models.yaml` has been changed externally.
This will reinitialize internal data structures
"""
# Reread models directory; note that this will reinitialize the cache,
# causing otherwise unreferenced models to be removed from memory
self._read_models()
def model_exists(
self,
model_name: str,
@ -382,6 +409,16 @@ class ModelManager(object):
def _get_model_cache_path(self, model_path):
return self.app_config.models_path / ".cache" / hashlib.md5(str(model_path).encode()).hexdigest()
@classmethod
def initialize_model_config(cls, config_path: Path):
"""Create empty config file"""
with open(config_path,'w') as yaml_file:
yaml_file.write(yaml.dump({'__metadata__':
{'version':'3.0.0'}
}
)
)
def get_model(
self,
model_name: str,
@ -404,7 +441,7 @@ class ModelManager(object):
if model_key not in self.models:
self.scan_models_directory(base_model=base_model, model_type=model_type)
if model_key not in self.models:
raise Exception(f"Model not found - {model_key}")
raise ModelNotFoundException(f"Model not found - {model_key}")
model_config = self.models[model_key]
model_path = self.app_config.root_path / model_config.path
@ -416,14 +453,14 @@ class ModelManager(object):
else:
self.models.pop(model_key, None)
raise Exception(f"Model not found - {model_key}")
raise ModelNotFoundException(f"Model not found - {model_key}")
# vae/movq override
# TODO:
if submodel_type is not None and hasattr(model_config, submodel_type):
override_path = getattr(model_config, submodel_type)
if override_path:
model_path = override_path
model_path = self.app_config.root_path / override_path
model_type = submodel_type
submodel_type = None
model_class = MODEL_CLASSES[base_model][model_type]
@ -431,6 +468,7 @@ class ModelManager(object):
# TODO: path
# TODO: is it accurate to use path as id
dst_convert_path = self._get_model_cache_path(model_path)
model_path = model_class.convert_if_required(
base_model=base_model,
model_path=str(model_path), # TODO: refactor str/Path types logic
@ -485,18 +523,36 @@ class ModelManager(object):
"""
return [(self.parse_key(x)) for x in self.models.keys()]
def list_model(
self,
model_name: str,
base_model: BaseModelType,
model_type: ModelType,
) -> dict:
"""
Returns a dict describing one installed model, using
the combined format of the list_models() method.
"""
models = self.list_models(base_model,model_type,model_name)
return models[0] if models else None
def list_models(
self,
base_model: Optional[BaseModelType] = None,
model_type: Optional[ModelType] = None,
model_name: Optional[str] = None,
) -> list[dict]:
"""
Return a list of models.
"""
model_keys = [self.create_key(model_name, base_model, model_type)] if model_name else sorted(self.models, key=str.casefold)
models = []
for model_key in sorted(self.models, key=str.casefold):
model_config = self.models[model_key]
for model_key in model_keys:
model_config = self.models.get(model_key)
if not model_config:
self.logger.error(f'Unknown model {model_name}')
raise ModelNotFoundException(f'Unknown model {model_name}')
cur_model_name, cur_base_model, cur_model_type = self.parse_key(model_key)
if base_model is not None and cur_base_model != base_model:
@ -507,9 +563,9 @@ class ModelManager(object):
model_dict = dict(
**model_config.dict(exclude_defaults=True),
# OpenAPIModelInfoBase
name=cur_model_name,
model_name=cur_model_name,
base_model=cur_base_model,
type=cur_model_type,
model_type=cur_model_type,
)
models.append(model_dict)
@ -540,10 +596,7 @@ class ModelManager(object):
model_cfg = self.models.pop(model_key, None)
if model_cfg is None:
self.logger.error(
f"Unknown model {model_key}"
)
return
raise ModelNotFoundException(f"Unknown model {model_key}")
# note: it not garantie to release memory(model can has other references)
cache_ids = self.cache_keys.pop(model_key, [])
@ -561,6 +614,7 @@ class ModelManager(object):
rmtree(str(model_path))
else:
model_path.unlink()
self.commit()
# LS: tested
def add_model(
@ -570,13 +624,16 @@ class ModelManager(object):
model_type: ModelType,
model_attributes: dict,
clobber: bool = False,
) -> None:
) -> AddModelResult:
"""
Update the named model with a dictionary of attributes. Will fail with an
assertion error if the name already exists. Pass clobber=True to overwrite.
On a successful update, the config will be changed in memory and the
method will return True. Will fail with an assertion error if provided
attributes are incorrect or the model name is missing.
The returned dict has the same format as the dict returned by
model_info().
"""
model_class = MODEL_CLASSES[base_model][model_type]
@ -600,13 +657,124 @@ class ModelManager(object):
old_model_cache.unlink()
# remove in-memory cache
# note: it not garantie to release memory(model can has other references)
# note: it not guaranteed to release memory(model can has other references)
cache_ids = self.cache_keys.pop(model_key, [])
for cache_id in cache_ids:
self.cache.uncache_model(cache_id)
self.models[model_key] = model_config
self.commit()
return AddModelResult(
name = model_name,
model_type = model_type,
base_model = base_model,
config = model_config,
)
def rename_model(
self,
model_name: str,
base_model: BaseModelType,
model_type: ModelType,
new_name: str = None,
new_base: BaseModelType = None,
):
'''
Rename or rebase a model.
'''
if new_name is None and new_base is None:
self.logger.error("rename_model() called with neither a new_name nor a new_base. {model_name} unchanged.")
return
model_key = self.create_key(model_name, base_model, model_type)
model_cfg = self.models.get(model_key, None)
if not model_cfg:
raise ModelNotFoundException(f"Unknown model: {model_key}")
old_path = self.app_config.root_path / model_cfg.path
new_name = new_name or model_name
new_base = new_base or base_model
new_key = self.create_key(new_name, new_base, model_type)
if new_key in self.models:
raise ValueError(f'Attempt to overwrite existing model definition "{new_key}"')
# if this is a model file/directory that we manage ourselves, we need to move it
if old_path.is_relative_to(self.app_config.models_path):
new_path = self.app_config.root_path / 'models' / new_base.value / model_type.value / new_name
move(old_path, new_path)
model_cfg.path = str(new_path.relative_to(self.app_config.root_path))
# clean up caches
old_model_cache = self._get_model_cache_path(old_path)
if old_model_cache.exists():
if old_model_cache.is_dir():
rmtree(str(old_model_cache))
else:
old_model_cache.unlink()
cache_ids = self.cache_keys.pop(model_key, [])
for cache_id in cache_ids:
self.cache.uncache_model(cache_id)
self.models.pop(model_key, None) # delete
self.models[new_key] = model_cfg
self.commit()
def convert_model (
self,
model_name: str,
base_model: BaseModelType,
model_type: Union[ModelType.Main,ModelType.Vae],
dest_directory: Optional[Path]=None,
) -> AddModelResult:
'''
Convert a checkpoint file into a diffusers folder, deleting the cached
version and deleting the original checkpoint file if it is in the models
directory.
:param model_name: Name of the model to convert
:param base_model: Base model type
:param model_type: Type of model ['vae' or 'main']
This will raise a ValueError unless the model is a checkpoint.
'''
info = self.model_info(model_name, base_model, model_type)
if info["model_format"] != "checkpoint":
raise ValueError(f"not a checkpoint format model: {model_name}")
# We are taking advantage of a side effect of get_model() that converts check points
# into cached diffusers directories stored at `location`. It doesn't matter
# what submodeltype we request here, so we get the smallest.
submodel = {"submodel_type": SubModelType.Tokenizer} if model_type==ModelType.Main else {}
model = self.get_model(model_name,
base_model,
model_type,
**submodel,
)
checkpoint_path = self.app_config.root_path / info["path"]
old_diffusers_path = self.app_config.models_path / model.location
new_diffusers_path = (dest_directory or self.app_config.models_path / base_model.value / model_type.value) / model_name
if new_diffusers_path.exists():
raise ValueError(f"A diffusers model already exists at {new_diffusers_path}")
try:
move(old_diffusers_path,new_diffusers_path)
info["model_format"] = "diffusers"
info["path"] = str(new_diffusers_path) if dest_directory else str(new_diffusers_path.relative_to(self.app_config.root_path))
info.pop('config')
result = self.add_model(model_name, base_model, model_type,
model_attributes = info,
clobber=True)
except:
# something went wrong, so don't leave dangling diffusers model in directory or it will cause a duplicate model error!
rmtree(new_diffusers_path)
raise
if checkpoint_path.exists() and checkpoint_path.is_relative_to(self.app_config.models_path):
checkpoint_path.unlink()
return result
def search_models(self, search_folder):
self.logger.info(f"Finding Models In: {search_folder}")
models_folder_ckpt = Path(search_folder).glob("**/*.ckpt")
@ -717,80 +885,68 @@ class ModelManager(object):
if model_path.is_relative_to(self.app_config.root_path):
model_path = model_path.relative_to(self.app_config.root_path)
try:
model_config: ModelConfigBase = model_class.probe_config(str(model_path))
self.models[model_key] = model_config
new_models_found = True
except NotImplementedError as e:
self.logger.warning(e)
try:
model_config: ModelConfigBase = model_class.probe_config(str(model_path))
self.models[model_key] = model_config
new_models_found = True
except InvalidModelException:
self.logger.warning(f"Not a valid model: {model_path}")
except NotImplementedError as e:
self.logger.warning(e)
imported_models = self.autoimport()
if (new_models_found or imported_models) and self.config_path:
self.commit()
def autoimport(self)->set[Path]:
def autoimport(self)->Dict[str, AddModelResult]:
'''
Scan the autoimport directory (if defined) and import new models, delete defunct models.
'''
# avoid circular import
from invokeai.backend.install.model_install_backend import ModelInstall
from invokeai.frontend.install.model_install import ask_user_for_prediction_type
class ScanAndImport(ModelSearch):
def __init__(self, directories, logger, ignore: Set[Path], installer: ModelInstall):
super().__init__(directories, logger)
self.installer = installer
self.ignore = ignore
def on_search_started(self):
self.new_models_found = dict()
def on_model_found(self, model: Path):
if model not in self.ignore:
self.new_models_found.update(self.installer.heuristic_import(model))
def on_search_completed(self):
self.logger.info(f'Scanned {self._items_scanned} files and directories, imported {len(self.new_models_found)} models')
def models_found(self):
return self.new_models_found
installer = ModelInstall(config = self.app_config,
model_manager = self,
prediction_type_helper = ask_user_for_prediction_type,
)
installed = set()
scanned_dirs = set()
config = self.app_config
known_paths = {(self.app_config.root_path / x['path']) for x in self.list_models()}
for autodir in [config.autoimport_dir,
config.lora_dir,
config.embedding_dir,
config.controlnet_dir]:
if autodir is None:
continue
self.logger.info(f'Scanning {autodir} for models to import')
autodir = self.app_config.root_path / autodir
if not autodir.exists():
continue
items_scanned = 0
new_models_found = set()
for root, dirs, files in os.walk(autodir):
items_scanned += len(dirs) + len(files)
for d in dirs:
path = Path(root) / d
if path in known_paths or path.parent in scanned_dirs:
scanned_dirs.add(path)
continue
if any([(path/x).exists() for x in {'config.json','model_index.json','learned_embeds.bin'}]):
new_models_found.update(installer.heuristic_install(path))
scanned_dirs.add(path)
for f in files:
path = Path(root) / f
if path in known_paths or path.parent in scanned_dirs:
continue
if path.suffix in {'.ckpt','.bin','.pth','.safetensors','.pt'}:
new_models_found.update(installer.heuristic_install(path))
self.logger.info(f'Scanned {items_scanned} files and directories, imported {len(new_models_found)} models')
installed.update(new_models_found)
return installed
known_paths = {config.root_path / x['path'] for x in self.list_models()}
directories = {config.root_path / x for x in [config.autoimport_dir,
config.lora_dir,
config.embedding_dir,
config.controlnet_dir]
}
scanner = ScanAndImport(directories, self.logger, ignore=known_paths, installer=installer)
scanner.search()
return scanner.models_found()
def heuristic_import(self,
items_to_import: Set[str],
prediction_type_helper: Callable[[Path],SchedulerPredictionType]=None,
)->Set[str]:
)->Dict[str, AddModelResult]:
'''Import a list of paths, repo_ids or URLs. Returns the set of
successfully imported items.
:param items_to_import: Set of strings corresponding to models to be imported.
@ -803,20 +959,24 @@ class ModelManager(object):
generally impossible to do this programmatically, so the
prediction_type_helper usually asks the user to choose.
The result is a set of successfully installed models. Each element
of the set is a dict corresponding to the newly-created OmegaConf stanza for
that model.
May return the following exceptions:
- ModelNotFoundException - one or more of the items to import is not a valid path, repo_id or URL
- ValueError - a corresponding model already exists
'''
# avoid circular import here
from invokeai.backend.install.model_install_backend import ModelInstall
successfully_installed = set()
successfully_installed = dict()
installer = ModelInstall(config = self.app_config,
prediction_type_helper = prediction_type_helper,
model_manager = self)
for thing in items_to_import:
try:
installed = installer.heuristic_install(thing)
successfully_installed.update(installed)
except Exception as e:
self.logger.warning(f'{thing} could not be imported: {str(e)}')
installed = installer.heuristic_import(thing)
successfully_installed.update(installed)
self.commit()
return successfully_installed

132
invokeai/backend/model_management/model_merge.py Normal file
View File

@ -0,0 +1,132 @@
"""
invokeai.backend.model_management.model_merge exports:
merge_diffusion_models() -- combine multiple models by location and return a pipeline object
merge_diffusion_models_and_commit() -- combine multiple models by ModelManager ID and write to models.yaml
Copyright (c) 2023 Lincoln Stein and the InvokeAI Development Team
"""
import warnings
from enum import Enum
from pathlib import Path
from diffusers import DiffusionPipeline
from diffusers import logging as dlogging
from typing import List, Union, Optional
import invokeai.backend.util.logging as logger
from ...backend.model_management import ModelManager, ModelType, BaseModelType, ModelVariantType, AddModelResult
class MergeInterpolationMethod(str, Enum):
WeightedSum = "weighted_sum"
Sigmoid = "sigmoid"
InvSigmoid = "inv_sigmoid"
AddDifference = "add_difference"
class ModelMerger(object):
def __init__(self, manager: ModelManager):
self.manager = manager
def merge_diffusion_models(
self,
model_paths: List[Path],
alpha: float = 0.5,
interp: MergeInterpolationMethod = None,
force: bool = False,
**kwargs,
) -> DiffusionPipeline:
"""
:param model_paths: up to three models, designated by their local paths or HuggingFace repo_ids
:param alpha: The interpolation parameter. Ranges from 0 to 1. It affects the ratio in which the checkpoints are merged. A 0.8 alpha
would mean that the first model checkpoints would affect the final result far less than an alpha of 0.2
:param interp: The interpolation method to use for the merging. Supports "sigmoid", "inv_sigmoid", "add_difference" and None.
Passing None uses the default interpolation which is weighted sum interpolation. For merging three checkpoints, only "add_difference" is supported.
:param force: Whether to ignore mismatch in model_config.json for the current models. Defaults to False.
**kwargs - the default DiffusionPipeline.get_config_dict kwargs:
cache_dir, resume_download, force_download, proxies, local_files_only, use_auth_token, revision, torch_dtype, device_map
"""
with warnings.catch_warnings():
warnings.simplefilter("ignore")
verbosity = dlogging.get_verbosity()
dlogging.set_verbosity_error()
pipe = DiffusionPipeline.from_pretrained(
model_paths[0],
custom_pipeline="checkpoint_merger",
)
merged_pipe = pipe.merge(
pretrained_model_name_or_path_list=model_paths,
alpha=alpha,
interp=interp.value if interp else None, #diffusers API treats None as "weighted sum"
force=force,
**kwargs,
)
dlogging.set_verbosity(verbosity)
return merged_pipe
def merge_diffusion_models_and_save (
self,
model_names: List[str],
base_model: Union[BaseModelType,str],
merged_model_name: str,
alpha: float = 0.5,
interp: MergeInterpolationMethod = None,
force: bool = False,
merge_dest_directory: Optional[Path] = None,
**kwargs,
) -> AddModelResult:
"""
:param models: up to three models, designated by their InvokeAI models.yaml model name
:param base_model: base model (must be the same for all merged models!)
:param merged_model_name: name for new model
:param alpha: The interpolation parameter. Ranges from 0 to 1. It affects the ratio in which the checkpoints are merged. A 0.8 alpha
would mean that the first model checkpoints would affect the final result far less than an alpha of 0.2
:param interp: The interpolation method to use for the merging. Supports "weighted_average", "sigmoid", "inv_sigmoid", "add_difference" and None.
Passing None uses the default interpolation which is weighted sum interpolation. For merging three checkpoints, only "add_difference" is supported. Add_difference is A+(B-C).
:param force: Whether to ignore mismatch in model_config.json for the current models. Defaults to False.
:param merge_dest_directory: Save the merged model to the designated directory (with 'merged_model_name' appended)
**kwargs - the default DiffusionPipeline.get_config_dict kwargs:
cache_dir, resume_download, force_download, proxies, local_files_only, use_auth_token, revision, torch_dtype, device_map
"""
model_paths = list()
config = self.manager.app_config
base_model = BaseModelType(base_model)
vae = None
for mod in model_names:
info = self.manager.list_model(mod, base_model=base_model, model_type=ModelType.Main)
assert info, f"model {mod}, base_model {base_model}, is unknown"
assert info["model_format"] == "diffusers", f"{mod} is not a diffusers model. It must be optimized before merging"
assert info["variant"] == "normal", f"{mod} is a {info['variant']} model, which cannot currently be merged"
assert len(model_names) <= 2 or \
interp==MergeInterpolationMethod.AddDifference, "When merging three models, only the 'add_difference' merge method is supported"
# pick up the first model's vae
if mod == model_names[0]:
vae = info.get("vae")
model_paths.extend([config.root_path / info["path"]])
merge_method = None if interp == 'weighted_sum' else MergeInterpolationMethod(interp)
logger.debug(f'interp = {interp}, merge_method={merge_method}')
merged_pipe = self.merge_diffusion_models(
model_paths, alpha, merge_method, force, **kwargs
)
dump_path = Path(merge_dest_directory) if merge_dest_directory else config.models_path / base_model.value / ModelType.Main.value
dump_path.mkdir(parents=True, exist_ok=True)
dump_path = dump_path / merged_model_name
merged_pipe.save_pretrained(dump_path, safe_serialization=1)
attributes = dict(
path = str(dump_path),
description = f"Merge of models {', '.join(model_names)}",
model_format = "diffusers",
variant = ModelVariantType.Normal.value,
vae = vae,
)
return self.manager.add_model(merged_model_name,
base_model = base_model,
model_type = ModelType.Main,
model_attributes = attributes,
clobber = True
)

Some files were not shown because too many files have changed in this diff Show More