fix(mm): fix update error

whoops!

.github/workflows/lint-frontend.yml

@@ -2,8 +2,6 @@ name: Lint frontend
 
 on:
   pull_request:
-    paths:
-      - 'invokeai/frontend/web/**'
     types:
       - 'ready_for_review'
       - 'opened'
@@ -11,8 +9,6 @@ on:
   push:
     branches:
       - 'main'
-    paths:
-      - 'invokeai/frontend/web/**'
   merge_group:
   workflow_dispatch:
 

.github/workflows/style-checks.yml

@@ -1,13 +1,14 @@
-name: Black  # TODO: add isort and flake8 later
+name: style checks
+# just formatting for now
+# TODO: add isort and flake8 later
 
 on:
-  pull_request: {}
+  pull_request:
   push:
-    branches: master
-    tags: "*"
+    branches: main
 
 jobs:
-  test:
+  black:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v3
@@ -19,8 +20,7 @@ jobs:
 
       - name: Install dependencies with pip
         run: |
-          pip install --upgrade pip wheel
-          pip install .[test]
+          pip install black
 
       # - run: isort --check-only .
       - run: black --check .

.github/workflows/test-invoke-pip-skip.yml

@@ -1,50 +0,0 @@
-name: Test invoke.py pip
-
-# This is a dummy stand-in for the actual tests
-# we don't need to run python tests on non-Python changes
-# But PRs require passing tests to be mergeable
-
-on:
-  pull_request:
-    paths:
-      - '**'
-      - '!pyproject.toml'
-      - '!invokeai/**'
-      - '!tests/**'
-      - 'invokeai/frontend/web/**'
-  merge_group:
-  workflow_dispatch:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  matrix:
-    if: github.event.pull_request.draft == false
-    strategy:
-      matrix:
-        python-version:
-          - '3.10'
-        pytorch:
-          - linux-cuda-11_7
-          - linux-rocm-5_2
-          - linux-cpu
-          - macos-default
-          - windows-cpu
-        include:
-          - pytorch: linux-cuda-11_7
-            os: ubuntu-22.04
-          - pytorch: linux-rocm-5_2
-            os: ubuntu-22.04
-          - pytorch: linux-cpu
-            os: ubuntu-22.04
-          - pytorch: macos-default
-            os: macOS-12
-          - pytorch: windows-cpu
-            os: windows-2022
-    name: ${{ matrix.pytorch }} on ${{ matrix.python-version }}
-    runs-on: ${{ matrix.os }}
-    steps:
-      - name: skip
-        run: echo "no build required"

.github/workflows/test-invoke-pip.yml

@@ -3,16 +3,7 @@ on:
   push:
     branches:
       - 'main'
-    paths:
-      - 'pyproject.toml'
-      - 'invokeai/**'
-      - '!invokeai/frontend/web/**'
   pull_request:
-    paths:
-      - 'pyproject.toml'
-      - 'invokeai/**'
-      - 'tests/**'
-      - '!invokeai/frontend/web/**'
     types:
       - 'ready_for_review'
       - 'opened'
@@ -65,10 +56,23 @@ jobs:
         id: checkout-sources
         uses: actions/checkout@v3
 
+      - name: Check for changed python files
+        id: changed-files
+        uses: tj-actions/changed-files@v37
+        with:
+          files_yaml: |
+            python:
+              - 'pyproject.toml'
+              - 'invokeai/**'
+              - '!invokeai/frontend/web/**'
+              - 'tests/**'
+
       - name: set test prompt to main branch validation
+        if: steps.changed-files.outputs.python_any_changed == 'true'
         run: echo "TEST_PROMPTS=tests/validate_pr_prompt.txt" >> ${{ matrix.github-env }}
 
       - name: setup python
+        if: steps.changed-files.outputs.python_any_changed == 'true'
         uses: actions/setup-python@v4
         with:
           python-version: ${{ matrix.python-version }}
@@ -76,6 +80,7 @@ jobs:
           cache-dependency-path: pyproject.toml
 
       - name: install invokeai
+        if: steps.changed-files.outputs.python_any_changed == 'true'
         env:
           PIP_EXTRA_INDEX_URL: ${{ matrix.extra-index-url }}
         run: >
@@ -83,6 +88,7 @@ jobs:
           --editable=".[test]"
 
       - name: run pytest
+        if: steps.changed-files.outputs.python_any_changed == 'true'
         id: run-pytest
         run: pytest
 

README.md

@@ -184,8 +184,9 @@ the command `npm install -g yarn` if needed)
 6. Configure InvokeAI and install a starting set of image generation models (you only need to do this once):
 
     ```terminal
-    invokeai-configure
+    invokeai-configure --root .
     ```
+	Don't miss the dot at the end!
 
 7. Launch the web server (do it every time you run InvokeAI):
 
@@ -193,15 +194,9 @@ the command `npm install -g yarn` if needed)
     invokeai-web
     ```
 
-8. Build Node.js assets
+8. Point your browser to http://localhost:9090 to bring up the web interface.
 
-  ```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`.
+9. 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`.

docs/contributing/contribution_guides/development.md

@@ -16,7 +16,7 @@ If you don't feel ready to make a code contribution yet, no problem! You can als
 There are two paths to making a development contribution: 
 
 1. Choosing an open issue to address. Open issues can be found in the [Issues](https://github.com/invoke-ai/InvokeAI/issues?q=is%3Aissue+is%3Aopen) section of the InvokeAI repository. These are tagged by the issue type (bug, enhancement, etc.) along with the “good first issues” tag denoting if they are suitable for first time contributors.
-    1. Additional items can be found on our roadmap <******************************link to roadmap>******************************. The roadmap is organized in terms of priority, and contains features of varying size and complexity. If there is an inflight item you’d like to help with, reach out to the contributor assigned to the item to see how you can help. 
+    1. Additional items can be found on our [roadmap](https://github.com/orgs/invoke-ai/projects/7). The roadmap is organized in terms of priority, and contains features of varying size and complexity. If there is an inflight item you’d like to help with, reach out to the contributor assigned to the item to see how you can help. 
 2. Opening a new issue or feature to add. **Please make sure you have searched through existing issues before creating new ones.**
 
 *Regardless of what you choose, please post in the  [#dev-chat](https://discord.com/channels/1020123559063990373/1049495067846524939) channel of the Discord before you start development in order to confirm that the issue or feature is aligned with the current direction of the project. We value our contributors time and effort and want to ensure that no one’s time is being misspent.*

docs/features/index.md

@@ -4,6 +4,9 @@ title: Overview
 
 Here you can find the documentation for InvokeAI's various features.
 
+## The [Getting Started Guide](../help/gettingStartedWithAI) 
+A getting started guide for those new to AI image generation. 
+
 ## The Basics
 ### * The [Web User Interface](WEB.md)
 Guide to the Web interface. Also see the [WebUI Hotkeys Reference Guide](WEBUIHOTKEYS.md)
@@ -46,7 +49,7 @@ Personalize models by adding your own style or subjects.
 
 ## Other Features
 
-### * [The NSFW Checker](NSFW.md)
+### * [The NSFW Checker](WATERMARK+NSFW.md)
 Prevent InvokeAI from displaying unwanted racy images.
 
 ### * [Controlling Logging](LOGGING.md)

docs/help/gettingStartedWithAI.md

@@ -0,0 +1,95 @@
+# Getting Started with AI Image Generation
+
+New to image generation with AI? You’re in the right place! 
+
+This is a high level walkthrough of some of the concepts and terms you’ll see as you start using InvokeAI. Please note, this is not an exhaustive guide and may be out of date due to the rapidly changing nature of the space. 
+
+## Using InvokeAI
+
+### **Prompt Crafting**
+
+- Prompts are the basis of using InvokeAI, providing the models directions on what to generate. As a general rule of thumb, the more detailed your prompt is, the better your result will be.
+
+ *To get started, here’s an easy template to use for structuring your prompts:*
+
+- Subject, Style, Quality, Aesthetic
+    - **Subject:** What your image will be about. E.g. “a futuristic city with trains”,  “penguins floating on icebergs”, “friends sharing beers”
+    - **Style:** The style or medium in which your image will be in. E.g. “photograph”, “pencil sketch”, “oil paints”, or “pop art”, “cubism”, “abstract”
+    - **Quality:** A particular aspect or trait that you would like to see emphasized in your image. E.g. "award-winning", "featured in {relevant set of high quality works}", "professionally acclaimed". Many people often use "masterpiece".
+    - **Aesthetics:** The visual impact and design of the artwork. This can be colors, mood, lighting, setting, etc.
+- There are two prompt boxes: *Positive Prompt* & *Negative Prompt*.
+    - A **Positive** Prompt includes words you want the model to reference when creating an image.
+    - Negative Prompt is for anything you want the model to eliminate when creating an image. It doesn’t always interpret things exactly the way you would, but helps control the generation process. Always try to include a few terms - you can typically use lower quality image terms like “blurry” or “distorted” with good success.
+- Some examples prompts you can try on your own:
+    - A detailed oil painting of a tranquil forest at sunset with vibrant+ colors and soft, golden light filtering through the trees
+    - friends sharing beers in a busy city, realistic colored pencil sketch, twilight, masterpiece, bright, lively
+
+### Generation Workflows
+
+- Invoke offers a number of different workflows for interacting with models to produce images. Each is extremely powerful on its own, but together provide you an unparalleled way of producing high quality creative outputs that align with your vision.
+    - **Text to Image:** The text to image tab focuses on the key workflow of using a prompt to generate a new image. It includes other features that help control the generation process as well.
+    - **Image to Image:** With image to image, you provide an image as a reference (called the “initial image”), which provides more guidance around color and structure to the AI as it generates a new image. This is provided alongside the same features as Text to Image.
+    - **Unified Canvas:** The Unified Canvas is an advanced AI-first image editing tool that is easy to use, but hard to master. Drag an image onto the canvas from your gallery in order to regenerate certain elements, edit content or colors (known as inpainting), or extend the image with an exceptional degree of consistency and clarity (called outpainting).
+
+### Improving Image Quality
+
+- Fine tuning your prompt - the more specific you are, the closer the image will turn out to what is in your head!  Adding more details in the Positive Prompt or Negative Prompt can help add / remove pieces of your image to improve it - You can also use advanced techniques like upweighting and downweighting to control the influence of certain words. [Learn more here](https://invoke-ai.github.io/InvokeAI/features/PROMPTS/#prompt-syntax-features).
+    - **Tip: If you’re seeing poor results, try adding the things you don’t like about the image to your negative prompt may help. E.g. distorted, low quality, unrealistic, etc.**
+- Explore different models - Other models can produce different results due to the data they’ve been trained on. Each model has specific language and settings it works best with; a model’s documentation is your friend here.  Play around with some and see what works best for you!
+- Increasing Steps - The number of steps used controls how much time the model is given to produce an image, and depends on the “Scheduler” used. The schedule controls how each step is processed by the model. More steps tends to mean better results, but will take longer - We recommend at least 30 steps for most
+- Tweak and Iterate - Remember, it’s best to change one thing at a time so you know what is working and what isn't. Sometimes you just need to try a new image, and other times using a new prompt might be the ticket. For testing, consider turning off the “random” Seed - Using the same seed with the same settings will produce the same image, which makes it the perfect way to learn exactly what your changes are doing.
+- Explore Advanced Settings - InvokeAI has a full suite of tools available to allow you complete control over your image creation process - Check out our [docs if you want to learn more](https://invoke-ai.github.io/InvokeAI/features/).
+
+
+## Terms & Concepts
+
+If you're interested in learning more, check out [this presentation](https://docs.google.com/presentation/d/1IO78i8oEXFTZ5peuHHYkVF-Y3e2M6iM5tCnc-YBfcCM/edit?usp=sharing) from one of our maintainers (@lstein). 
+
+### Stable Diffusion
+
+Stable Diffusion is deep learning, text-to-image model that is the foundation of the capabilities found in InvokeAI. Since the release of Stable Diffusion, there have been many subsequent models created based on Stable Diffusion that are designed to generate specific types of images. 
+
+### Prompts
+
+Prompts provide the models directions on what to generate. As a general rule of thumb, the more detailed your prompt is, the better your result will be.
+
+### Models
+
+Models are the magic that power InvokeAI. These files represent the output of training a machine on understanding massive amounts of images - providing them with the capability to generate new images using just a text description of what you’d like to see. (Like Stable Diffusion!)
+
+Invoke offers a simple way to download several different models upon installation, but many more can be discovered online, including at ****. Each model can produce a unique style of output, based on the images it was trained on - Try out different models to see which best fits your creative vision!
+
+- *Models that contain “inpainting” in the name are designed for use with the inpainting feature of the Unified Canvas*
+
+### Scheduler
+
+Schedulers guide the process of removing noise (de-noising) from data. They determine:
+
+1. The number of steps to take to remove the noise.
+2. Whether the steps are random (stochastic) or predictable (deterministic).
+3. The specific method (algorithm) used for de-noising.
+
+Experimenting with different schedulers is recommended as each will produce different outputs!
+
+### Steps
+
+The number of de-noising steps each generation through. 
+
+Schedulers can be intricate and there's often a balance to strike between how quickly they can de-noise data and how well they can do it. It's typically advised to experiment with different schedulers to see which one gives the best results. There has been a lot written on the internet about different schedulers, as well as exploring what the right level of "steps" are for each. You can save generation time by reducing the number of steps used, but you'll want to make sure that you are satisfied with the quality of images produced!
+
+### Low-Rank Adaptations / LoRAs
+
+Low-Rank Adaptations (LoRAs) are like a smaller, more focused version of models, intended to focus on training a better understanding of how a specific character, style, or concept looks.
+
+### Textual Inversion Embeddings
+
+Textual Inversion Embeddings, like LoRAs, assist with more easily prompting for certain characters, styles, or concepts. However, embeddings are trained to update the relationship between a specific word (known as the “trigger”) and the intended output. 
+
+### ControlNet
+
+ControlNets are neural network models that are able to extract key features from an existing image and use these features to guide the output of the image generation model. 
+
+### VAE
+
+Variational auto-encoder (VAE) is a encode/decode model that translates the "latents" image produced during the image generation procees to the large pixel images that we see. 
+

docs/index.md

@@ -11,6 +11,33 @@ title: Home
   ```
 -->
 
+<!-- CSS styling -->
+<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/@fortawesome/fontawesome-free@6.2.1/css/fontawesome.min.css">
+<style>
+    .button {
+      width: 300px;
+      height: 50px;
+      background-color: #448AFF;
+      color: #fff;
+      font-size: 16px;
+      border: none;
+      cursor: pointer;
+      border-radius: 0.2rem;
+    }
+
+    .button-container {
+      display: grid;
+      grid-template-columns: repeat(3, 300px);
+      gap: 20px;
+    }
+
+    .button:hover {
+      background-color: #526CFE;
+    }
+</style>
+
+
+
 <div align="center" markdown>
 
 
@@ -70,63 +97,23 @@ image-to-image generator. It provides a streamlined process with various new
 features and options to aid the image generation process. It runs on Windows,
 Mac and Linux machines, and runs on GPU cards with as little as 4 GB of RAM.
 
-**Quick links**: [<a href="https://discord.gg/ZmtBAhwWhy">Discord Server</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"><img src="assets/invoke-web-server-1.png" width=640></div>
 
-!!! note
+!!! Note
 
-    This software is rapidly evolving. Please use the [Issues tab](https://github.com/invoke-ai/InvokeAI/issues) to report bugs and make feature requests. Be sure to use the provided templates. They will help aid diagnose issues faster.
+    This project is rapidly evolving. Please use the [Issues tab](https://github.com/invoke-ai/InvokeAI/issues) to report bugs and make feature requests. Be sure to use the provided templates as it will help aid response time.
 
-## :octicons-package-dependencies-24: Installation
+## :octicons-link-24: Quick Links
 
-This software is supported across Linux, Windows and Macintosh. Linux users can use
-either an Nvidia-based card (with CUDA support) or an AMD card (using the ROCm
-driver).
-
-### [Installation Getting Started Guide](installation)
-#### **[Automated Installer](installation/010_INSTALL_AUTOMATED.md)**
-✅ This is the recommended installation method for first-time users. 
-#### [Manual Installation](installation/020_INSTALL_MANUAL.md)
-This method is recommended for experienced users and developers
-#### [Docker Installation](installation/040_INSTALL_DOCKER.md)
-This method is recommended for those familiar with running Docker containers
-#### [Installation Troubleshooting](installation/010_INSTALL_AUTOMATED.md#troubleshooting)
-Installation troubleshooting guide.
-### Other Installation Guides
-  - [PyPatchMatch](installation/060_INSTALL_PATCHMATCH.md)
-  - [XFormers](installation/070_INSTALL_XFORMERS.md)
-  - [CUDA and ROCm Drivers](installation/030_INSTALL_CUDA_AND_ROCM.md)
-  - [Installing New Models](installation/050_INSTALLING_MODELS.md)
-
-## :fontawesome-solid-computer: Hardware Requirements
-
-### :octicons-cpu-24: System
-
-You wil need one of the following:
-
-- :simple-nvidia: An NVIDIA-based graphics card with 4 GB or more VRAM memory.
-- :simple-amd: An AMD-based graphics card with 4 GB or more VRAM memory (Linux
-  only)
-- :fontawesome-brands-apple: An Apple computer with an M1 chip.
-
-We do **not recommend** the following video cards due to issues with their
-running in half-precision mode and having insufficient VRAM to render 512x512
-images in full-precision mode:
-
-- NVIDIA 10xx series cards such as the 1080ti
-- GTX 1650 series cards
-- GTX 1660 series cards
-
-### :fontawesome-solid-memory: Memory and Disk
-
-- At least 12 GB Main Memory RAM.
-- At least 18 GB of free disk space for the machine learning model, Python, and
-  all its dependencies.
+<div class="button-container">
+    <a href="installation/INSTALLATION"> <button class="button">Installation</button> </a>
+    <a href="features/"> <button class="button">Features</button> </a>
+    <a href="help/gettingStartedWithAI/"> <button class="button">Getting Started</button> </a>
+    <a href="contributing/CONTRIBUTING/"> <button class="button">Contributing</button> </a>
+    <a href="https://github.com/invoke-ai/InvokeAI/"> <button class="button">Code and Downloads</button> </a>
+    <a href="https://github.com/invoke-ai/InvokeAI/issues"> <button class="button">Bug Reports </button> </a>
+    <a href="https://discord.gg/ZmtBAhwWhy"> <button class="button"> Join the Discord Server!</button> </a>
+</div>
 
 
 ## :octicons-gift-24: InvokeAI Features

docs/installation/010_INSTALL_AUTOMATED.md

@@ -394,7 +394,7 @@ rm .\.venv -r -force
 python -mvenv .venv
 .\.venv\Scripts\activate
 pip install invokeai
-invokeai-configure --root .
+invokeai-configure --yes --root .
 ```
 
 If you see anything marked as an error during this process please stop

docs/installation/020_INSTALL_MANUAL.md

@@ -192,8 +192,10 @@ manager, please follow these steps:
     your outputs.
 
     ```terminal
-    invokeai-configure
+    invokeai-configure --root .
     ```
+	
+	Don't miss the dot at the end of the command!
 
     The script `invokeai-configure` will interactively guide you through the
     process of downloading and installing the weights files needed for InvokeAI.
@@ -225,12 +227,6 @@ manager, please follow these steps:
 
         !!! warning "Make sure that the virtual environment is activated, which should create `(.venv)` in front of your prompt!"
 
-        === "CLI"
-
-            ```bash
-            invokeai
-            ```
-
         === "local Webserver"
 
             ```bash
@@ -243,6 +239,12 @@ manager, please follow these steps:
             invokeai --web --host 0.0.0.0
             ```
 
+        === "CLI"
+
+            ```bash
+            invokeai
+            ```
+
         If you choose the run the web interface, point your browser at
         http://localhost:9090 in order to load the GUI.
 

docs/installation/050_INSTALLING_MODELS.md

@@ -124,7 +124,7 @@ installation. Examples:
 invokeai-model-install --list controlnet
 
 # (install the model at the indicated URL)
-invokeai-model-install --add http://civitai.com/2860
+invokeai-model-install --add https://civitai.com/api/download/models/128713
 
 # (delete the named model)
 invokeai-model-install --delete sd-1/main/analog-diffusion
@@ -170,4 +170,4 @@ elsewhere on disk and they will be autoimported. You can also create
 subfolders and organize them as you wish.
 
 The location of the autoimport directories are controlled by settings
-in `invokeai.yaml`. See [Configuration](../features/CONFIGURATION.md).
+in `invokeai.yaml`. See [Configuration](../features/CONFIGURATION.md).

docs/installation/INSTALLATION.md

@@ -1,6 +1,4 @@
----
-title: Overview
----
+# Overview
 
 We offer several ways to install InvokeAI, each one suited to your
 experience and preferences. We suggest that everyone start by
@@ -15,6 +13,56 @@ See the [troubleshooting
 section](010_INSTALL_AUTOMATED.md#troubleshooting) of the automated
 install guide for frequently-encountered installation issues.
 
+This fork is supported across Linux, Windows and Macintosh. Linux users can use
+either an Nvidia-based card (with CUDA support) or an AMD card (using the ROCm
+driver).
+
+### [Installation Getting Started Guide](installation)
+#### **[Automated Installer](010_INSTALL_AUTOMATED.md)**
+✅ This is the recommended installation method for first-time users. 
+#### [Manual Installation](020_INSTALL_MANUAL.md)
+This method is recommended for experienced users and developers
+#### [Docker Installation](040_INSTALL_DOCKER.md)
+This method is recommended for those familiar with running Docker containers
+### Other Installation Guides
+  - [PyPatchMatch](installation/060_INSTALL_PATCHMATCH.md)
+  - [XFormers](installation/070_INSTALL_XFORMERS.md)
+  - [CUDA and ROCm Drivers](installation/030_INSTALL_CUDA_AND_ROCM.md)
+  - [Installing New Models](installation/050_INSTALLING_MODELS.md)
+
+## :fontawesome-solid-computer: Hardware Requirements
+
+### :octicons-cpu-24: System
+
+You wil need one of the following:
+
+- :simple-nvidia: An NVIDIA-based graphics card with 4 GB or more VRAM memory.
+- :simple-amd: An AMD-based graphics card with 4 GB or more VRAM memory (Linux
+  only)
+- :fontawesome-brands-apple: An Apple computer with an M1 chip.
+
+** SDXL 1.0 Requirements*
+To use SDXL, user must have one of the following: 
+- :simple-nvidia: An NVIDIA-based graphics card with 8 GB or more VRAM memory.
+- :simple-amd: An AMD-based graphics card with 16 GB or more VRAM memory (Linux
+  only)
+- :fontawesome-brands-apple: An Apple computer with an M1 chip.
+
+
+### :fontawesome-solid-memory: Memory and Disk
+
+- At least 12 GB Main Memory RAM.
+- At least 18 GB of free disk space for the machine learning model, Python, and
+  all its dependencies.
+
+We do **not recommend** the following video cards due to issues with their
+running in half-precision mode and having insufficient VRAM to render 512x512
+images in full-precision mode:
+
+- NVIDIA 10xx series cards such as the 1080ti
+- GTX 1650 series cards
+- GTX 1660 series cards
+
 ## Installation options
 
 1. [Automated Installer](010_INSTALL_AUTOMATED.md)

docs/nodes/communityNodes.md

@@ -14,23 +14,28 @@ The nodes linked below have been developed and contributed by members of the Inv
 
 ## List of Nodes
 
-### Face Mask
+### FaceTools
 
-**Description:** This node autodetects a face in the image using MediaPipe and masks it by making it transparent. Via outpainting you can swap faces with other faces, or invert the mask and swap things around the face with other things. Additionally, you can supply X and Y offset values to scale/change the shape of the mask for finer control. The node also outputs an all-white mask in the same dimensions as the input image. This is needed by the inpaint node (and unified canvas) for outpainting.
+**Description:** FaceTools is a collection of nodes created to manipulate faces as you would in Unified Canvas. It includes FaceMask, FaceOff, and FacePlace. FaceMask autodetects a face in the image using MediaPipe and creates a mask from it. FaceOff similarly detects a face, then takes the face off of the image by adding a square bounding box around it and cropping/scaling it. FacePlace puts the bounded face image from FaceOff back onto the original image. Using these nodes with other inpainting node(s), you can put new faces on existing things, put new things around existing faces, and work closer with a face as a bounded image. Additionally, you can supply X and Y offset values to scale/change the shape of the mask for finer control on FaceMask and FaceOff. See GitHub repository below for usage examples.
 
-**Node Link:** https://github.com/ymgenesis/InvokeAI/blob/facemaskmediapipe/invokeai/app/invocations/facemask.py
+**Node Link:** https://github.com/ymgenesis/FaceTools/
 
-**Example Node Graph:** https://www.mediafire.com/file/gohn5sb1bfp8use/21-July_2023-FaceMask.json/file
+**FaceMask Output Examples** 
 
-**Output Examples** 
+![5cc8abce-53b0-487a-b891-3bf94dcc8960](https://github.com/invoke-ai/InvokeAI/assets/25252829/43f36d24-1429-4ab1-bd06-a4bedfe0955e)
+![b920b710-1882-49a0-8d02-82dff2cca907](https://github.com/invoke-ai/InvokeAI/assets/25252829/7660c1ed-bf7d-4d0a-947f-1fc1679557ba)
+![71a91805-fda5-481c-b380-264665703133](https://github.com/invoke-ai/InvokeAI/assets/25252829/f8f6a2ee-2b68-4482-87da-b90221d5c3e2)
 
-![2e3168cb-af6a-475d-bfac-c7b7fd67b4c2](https://github.com/ymgenesis/InvokeAI/assets/25252829/a5ad7d44-2ada-4b3c-a56e-a21f8244a1ac)
-![2_annotated](https://github.com/ymgenesis/InvokeAI/assets/25252829/53416c8a-a23b-4d76-bb6d-3cfd776e0096)
-![2fe2150c-fd08-4e26-8c36-f0610bf441bb](https://github.com/ymgenesis/InvokeAI/assets/25252829/b0f7ecfe-f093-4147-a904-b9f131b41dc9)
-![831b6b98-4f0f-4360-93c8-69a9c1338cbe](https://github.com/ymgenesis/InvokeAI/assets/25252829/fc7b0622-e361-4155-8a76-082894d084f0)
+<hr>
+
+### Ideal Size
+
+**Description:** This node calculates an ideal image size for a first pass of a multi-pass upscaling. The aim is to avoid duplication that results from choosing a size larger than the model is capable of.
+
+**Node Link:** https://github.com/JPPhoto/ideal-size-node
 
 --------------------------------
-### Super Cool Node Template
+### Example Node Template
 
 **Description:** This node allows you to do super cool things with InvokeAI.
 
@@ -40,13 +45,9 @@ The nodes linked below have been developed and contributed by members of the Inv
 
 **Output Examples** 
 
-![Invoke AI](https://invoke-ai.github.io/InvokeAI/assets/invoke_ai_banner.png)
-
-### Ideal Size
-
-**Description:** This node calculates an ideal image size for a first pass of a multi-pass upscaling. The aim is to avoid duplication that results from choosing a size larger than the model is capable of.
-
-**Node Link:** https://github.com/JPPhoto/ideal-size-node
+![Example Image](https://invoke-ai.github.io/InvokeAI/assets/invoke_ai_banner.png){: style="height:115px;width:240px"}
 
 ## Help
 If you run into any issues with a node, please post in the [InvokeAI Discord](https://discord.gg/ZmtBAhwWhy). 
+
+

flake.lock

@@ -0,0 +1,25 @@
+{
+  "nodes": {
+    "nixpkgs": {
+      "locked": {
+        "lastModified": 1690630721,
+        "narHash": "sha256-Y04onHyBQT4Erfr2fc82dbJTfXGYrf4V0ysLUYnPOP8=",
+        "owner": "NixOS",
+        "repo": "nixpkgs",
+        "rev": "d2b52322f35597c62abf56de91b0236746b2a03d",
+        "type": "github"
+      },
+      "original": {
+        "id": "nixpkgs",
+        "type": "indirect"
+      }
+    },
+    "root": {
+      "inputs": {
+        "nixpkgs": "nixpkgs"
+      }
+    }
+  },
+  "root": "root",
+  "version": 7
+}

flake.nix

@@ -0,0 +1,91 @@
+# Important note: this flake does not attempt to create a fully isolated, 'pure'
+# Python environment for InvokeAI. Instead, it depends on local invocations of
+# virtualenv/pip to install the required (binary) packages, most importantly the
+# prebuilt binary pytorch packages with CUDA support.
+# ML Python packages with CUDA support, like pytorch, are notoriously expensive
+# to compile so it's purposefuly not what this flake does.
+
+{
+  description = "An (impure) flake to develop on InvokeAI.";
+
+  outputs = { self, nixpkgs }:
+    let
+      system = "x86_64-linux";
+      pkgs = import nixpkgs {
+        inherit system;
+        config.allowUnfree = true;
+      };
+
+      python = pkgs.python310;
+
+      mkShell = { dir, install }:
+        let
+          setupScript = pkgs.writeScript "setup-invokai" ''
+            # This must be sourced using 'source', not executed.
+            ${python}/bin/python -m venv ${dir}
+            ${dir}/bin/python -m pip install ${install}
+            # ${dir}/bin/python -c 'import torch; assert(torch.cuda.is_available())'
+            source ${dir}/bin/activate
+          '';
+        in
+        pkgs.mkShell rec {
+          buildInputs = with pkgs; [
+            # Backend: graphics, CUDA.
+            cudaPackages.cudnn
+            cudaPackages.cuda_nvrtc
+            cudatoolkit
+            pkgconfig
+            libconfig
+            cmake
+            blas
+            freeglut
+            glib
+            gperf
+            procps
+            libGL
+            libGLU
+            linuxPackages.nvidia_x11
+            python
+            (opencv4.override {
+              enableGtk3 = true;
+              enableFfmpeg = true;
+              enableCuda = true;
+              enableUnfree = true;
+            })
+            stdenv.cc
+            stdenv.cc.cc.lib
+            xorg.libX11
+            xorg.libXext
+            xorg.libXi
+            xorg.libXmu
+            xorg.libXrandr
+            xorg.libXv
+            zlib
+
+            # Pre-commit hooks.
+            black
+
+            # Frontend.
+            yarn
+            nodejs
+          ];
+          LD_LIBRARY_PATH = pkgs.lib.makeLibraryPath buildInputs;
+          CUDA_PATH = pkgs.cudatoolkit;
+          EXTRA_LDFLAGS = "-L${pkgs.linuxPackages.nvidia_x11}/lib";
+          shellHook = ''
+            if [[ -f "${dir}/bin/activate" ]]; then
+              source "${dir}/bin/activate"
+              echo "Using Python: $(which python)"
+            else
+              echo "Use 'source ${setupScript}' to set up the environment."
+            fi
+          '';
+        };
+    in
+    {
+      devShells.${system} = rec {
+        develop = mkShell { dir = "venv"; install = "-e '.[xformers]' --extra-index-url https://download.pytorch.org/whl/cu118"; };
+        default = develop;
+      };
+    };
+}

installer/lib/installer.py

@@ -13,7 +13,7 @@ from pathlib import Path
 from tempfile import TemporaryDirectory
 from typing import Union
 
-SUPPORTED_PYTHON = ">=3.9.0,<3.11"
+SUPPORTED_PYTHON = ">=3.9.0,<=3.11.100"
 INSTALLER_REQS = ["rich", "semver", "requests", "plumbum", "prompt-toolkit"]
 BOOTSTRAP_VENV_PREFIX = "invokeai-installer-tmp"
 
@@ -149,7 +149,7 @@ class Installer:
         return venv_dir
 
     def install(
-        self, root: str = "~/invokeai-3", version: str = "latest", yes_to_all=False, find_links: Path = None
+        self, root: str = "~/invokeai", version: str = "latest", yes_to_all=False, find_links: Path = None
     ) -> None:
         """
         Install the InvokeAI application into the given runtime path
@@ -168,7 +168,8 @@ class Installer:
 
         messages.welcome()
 
-        self.dest = Path(root).expanduser().resolve() if yes_to_all else messages.dest_path(root)
+        default_path = os.environ.get("INVOKEAI_ROOT") or Path(root).expanduser().resolve()
+        self.dest = default_path if yes_to_all else messages.dest_path(root)
 
         # create the venv for the app
         self.venv = self.app_venv()
@@ -248,6 +249,9 @@ class InvokeAiInstance:
             pip[
                 "install",
                 "--require-virtualenv",
+                "numpy~=1.24.0",  # choose versions that won't be uninstalled during phase 2
+                "urllib3~=1.26.0",
+                "requests~=2.28.0",
                 "torch~=2.0.0",
                 "torchmetrics==0.11.4",
                 "torchvision>=0.14.1",
@@ -451,7 +455,7 @@ def get_torch_source() -> (Union[str, None], str):
     device = graphical_accelerator()
 
     url = None
-    optional_modules = None
+    optional_modules = "[onnx]"
     if OS == "Linux":
         if device == "rocm":
             url = "https://download.pytorch.org/whl/rocm5.4.2"
@@ -460,7 +464,10 @@ def get_torch_source() -> (Union[str, None], str):
 
     if device == "cuda":
         url = "https://download.pytorch.org/whl/cu117"
-        optional_modules = "[xformers]"
+        optional_modules = "[xformers,onnx-cuda]"
+    if device == "cuda_and_dml":
+        url = "https://download.pytorch.org/whl/cu117"
+        optional_modules = "[xformers,onnx-directml]"
 
     # in all other cases, Torch wheels should be coming from PyPi as of Torch 1.13
 

installer/lib/main.py

@@ -3,6 +3,7 @@ InvokeAI Installer
 """
 
 import argparse
+import os
 from pathlib import Path
 from installer import Installer
 
@@ -15,7 +16,7 @@ if __name__ == "__main__":
         dest="root",
         type=str,
         help="Destination path for installation",
-        default="~/invokeai",
+        default=os.environ.get("INVOKEAI_ROOT") or "~/invokeai",
     )
     parser.add_argument(
         "-y",

installer/lib/messages.py

@@ -167,6 +167,10 @@ def graphical_accelerator():
         "an [gold1 b]NVIDIA[/] GPU (using CUDA™)",
         "cuda",
     )
+    nvidia_with_dml = (
+        "an [gold1 b]NVIDIA[/] GPU (using CUDA™, and DirectML™ for ONNX) -- ALPHA",
+        "cuda_and_dml",
+    )
     amd = (
         "an [gold1 b]AMD[/] GPU (using ROCm™)",
         "rocm",
@@ -181,7 +185,7 @@ def graphical_accelerator():
     )
 
     if OS == "Windows":
-        options = [nvidia, cpu]
+        options = [nvidia, nvidia_with_dml, cpu]
     if OS == "Linux":
         options = [nvidia, amd, cpu]
     elif OS == "Darwin":

installer/templates/invoke.bat.in

@@ -41,7 +41,7 @@ IF /I "%choice%" == "1" (
     python .venv\Scripts\invokeai-configure.exe --skip-sd-weight --skip-support-models
 ) ELSE IF /I "%choice%" == "7" (
     echo Running invokeai-configure...
-    python .venv\Scripts\invokeai-configure.exe --yes --default_only
+    python .venv\Scripts\invokeai-configure.exe --yes --skip-sd-weight
 ) ELSE IF /I "%choice%" == "8" (
     echo Developer Console
     echo Python command is:

installer/templates/invoke.sh.in

@@ -82,7 +82,7 @@ do_choice() {
     7)
         clear
         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
+        invokeai-configure --root ${INVOKEAI_ROOT} --yes --default_only --skip-sd-weights
         ;;
     8)
         clear

invokeai/app/api/dependencies.py

@@ -1,7 +1,7 @@
 # Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
 
+from typing import Optional
 from logging import Logger
-import os
 from invokeai.app.services.board_image_record_storage import (
     SqliteBoardImageRecordStorage,
 )
@@ -29,6 +29,7 @@ from ..services.invoker import Invoker
 from ..services.processor import DefaultInvocationProcessor
 from ..services.sqlite import SqliteItemStorage
 from ..services.model_manager_service import ModelManagerService
+from ..services.invocation_stats import InvocationStatsService
 from .events import FastAPIEventService
 
 
@@ -54,7 +55,7 @@ logger = InvokeAILogger.getLogger()
 class ApiDependencies:
     """Contains and initializes all dependencies for the API"""
 
-    invoker: Invoker = None
+    invoker: Invoker
 
     @staticmethod
     def initialize(config: InvokeAIAppConfig, event_handler_id: int, logger: Logger = logger):
@@ -67,8 +68,9 @@ class ApiDependencies:
         output_folder = config.output_path
 
         # TODO: build a file/path manager?
-        db_location = config.db_path
-        db_location.parent.mkdir(parents=True, exist_ok=True)
+        db_path = config.db_path
+        db_path.parent.mkdir(parents=True, exist_ok=True)
+        db_location = str(db_path)
 
         graph_execution_manager = SqliteItemStorage[GraphExecutionState](
             filename=db_location, table_name="graph_executions"
@@ -127,6 +129,7 @@ class ApiDependencies:
             graph_execution_manager=graph_execution_manager,
             processor=DefaultInvocationProcessor(),
             configuration=config,
+            performance_statistics=InvocationStatsService(graph_execution_manager),
             logger=logger,
         )
 

invokeai/app/api/routers/board_images.py

@@ -1,24 +1,30 @@
-from fastapi import Body, HTTPException, Path, Query
+from fastapi import Body, HTTPException
 from fastapi.routing import APIRouter
-from invokeai.app.services.board_record_storage import BoardRecord, BoardChanges
-from invokeai.app.services.image_record_storage import OffsetPaginatedResults
-from invokeai.app.services.models.board_record import BoardDTO
-from invokeai.app.services.models.image_record import ImageDTO
+from pydantic import BaseModel, Field
 
 from ..dependencies import ApiDependencies
 
 board_images_router = APIRouter(prefix="/v1/board_images", tags=["boards"])
 
 
+class AddImagesToBoardResult(BaseModel):
+    board_id: str = Field(description="The id of the board the images were added to")
+    added_image_names: list[str] = Field(description="The image names that were added to the board")
+
+
+class RemoveImagesFromBoardResult(BaseModel):
+    removed_image_names: list[str] = Field(description="The image names that were removed from their board")
+
+
 @board_images_router.post(
     "/",
-    operation_id="create_board_image",
+    operation_id="add_image_to_board",
     responses={
         201: {"description": "The image was added to a board successfully"},
     },
     status_code=201,
 )
-async def create_board_image(
+async def add_image_to_board(
     board_id: str = Body(description="The id of the board to add to"),
     image_name: str = Body(description="The name of the image to add"),
 ):
@@ -29,26 +35,78 @@ async def create_board_image(
         )
         return result
     except Exception as e:
-        raise HTTPException(status_code=500, detail="Failed to add to board")
+        raise HTTPException(status_code=500, detail="Failed to add image to board")
 
 
 @board_images_router.delete(
     "/",
-    operation_id="remove_board_image",
+    operation_id="remove_image_from_board",
     responses={
         201: {"description": "The image was removed from the board successfully"},
     },
     status_code=201,
 )
-async def remove_board_image(
-    board_id: str = Body(description="The id of the board"),
-    image_name: str = Body(description="The name of the image to remove"),
+async def remove_image_from_board(
+    image_name: str = Body(description="The name of the image to remove", embed=True),
 ):
-    """Deletes a board_image"""
+    """Removes an image from its board, if it had one"""
     try:
-        result = ApiDependencies.invoker.services.board_images.remove_image_from_board(
-            board_id=board_id, image_name=image_name
-        )
+        result = ApiDependencies.invoker.services.board_images.remove_image_from_board(image_name=image_name)
         return result
     except Exception as e:
-        raise HTTPException(status_code=500, detail="Failed to update board")
+        raise HTTPException(status_code=500, detail="Failed to remove image from board")
+
+
+@board_images_router.post(
+    "/batch",
+    operation_id="add_images_to_board",
+    responses={
+        201: {"description": "Images were added to board successfully"},
+    },
+    status_code=201,
+    response_model=AddImagesToBoardResult,
+)
+async def add_images_to_board(
+    board_id: str = Body(description="The id of the board to add to"),
+    image_names: list[str] = Body(description="The names of the images to add", embed=True),
+) -> AddImagesToBoardResult:
+    """Adds a list of images to a board"""
+    try:
+        added_image_names: list[str] = []
+        for image_name in image_names:
+            try:
+                ApiDependencies.invoker.services.board_images.add_image_to_board(
+                    board_id=board_id, image_name=image_name
+                )
+                added_image_names.append(image_name)
+            except:
+                pass
+        return AddImagesToBoardResult(board_id=board_id, added_image_names=added_image_names)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail="Failed to add images to board")
+
+
+@board_images_router.post(
+    "/batch/delete",
+    operation_id="remove_images_from_board",
+    responses={
+        201: {"description": "Images were removed from board successfully"},
+    },
+    status_code=201,
+    response_model=RemoveImagesFromBoardResult,
+)
+async def remove_images_from_board(
+    image_names: list[str] = Body(description="The names of the images to remove", embed=True),
+) -> RemoveImagesFromBoardResult:
+    """Removes a list of images from their board, if they had one"""
+    try:
+        removed_image_names: list[str] = []
+        for image_name in image_names:
+            try:
+                ApiDependencies.invoker.services.board_images.remove_image_from_board(image_name=image_name)
+                removed_image_names.append(image_name)
+            except:
+                pass
+        return RemoveImagesFromBoardResult(removed_image_names=removed_image_names)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail="Failed to remove images from board")

invokeai/app/api/routers/images.py

@@ -5,6 +5,7 @@ from fastapi import Body, HTTPException, Path, Query, Request, Response, UploadF
 from fastapi.responses import FileResponse
 from fastapi.routing import APIRouter
 from PIL import Image
+from pydantic import BaseModel, Field
 
 from invokeai.app.invocations.metadata import ImageMetadata
 from invokeai.app.models.image import ImageCategory, ResourceOrigin
@@ -25,7 +26,7 @@ IMAGE_MAX_AGE = 31536000
 
 
 @images_router.post(
-    "/",
+    "/upload",
     operation_id="upload_image",
     responses={
         201: {"description": "The image was uploaded successfully"},
@@ -77,7 +78,7 @@ async def upload_image(
         raise HTTPException(status_code=500, detail="Failed to create image")
 
 
-@images_router.delete("/{image_name}", operation_id="delete_image")
+@images_router.delete("/i/{image_name}", operation_id="delete_image")
 async def delete_image(
     image_name: str = Path(description="The name of the image to delete"),
 ) -> None:
@@ -103,7 +104,7 @@ async def clear_intermediates() -> int:
 
 
 @images_router.patch(
-    "/{image_name}",
+    "/i/{image_name}",
     operation_id="update_image",
     response_model=ImageDTO,
 )
@@ -120,7 +121,7 @@ async def update_image(
 
 
 @images_router.get(
-    "/{image_name}",
+    "/i/{image_name}",
     operation_id="get_image_dto",
     response_model=ImageDTO,
 )
@@ -136,7 +137,7 @@ async def get_image_dto(
 
 
 @images_router.get(
-    "/{image_name}/metadata",
+    "/i/{image_name}/metadata",
     operation_id="get_image_metadata",
     response_model=ImageMetadata,
 )
@@ -152,7 +153,7 @@ async def get_image_metadata(
 
 
 @images_router.get(
-    "/{image_name}/full",
+    "/i/{image_name}/full",
     operation_id="get_image_full",
     response_class=Response,
     responses={
@@ -187,7 +188,7 @@ async def get_image_full(
 
 
 @images_router.get(
-    "/{image_name}/thumbnail",
+    "/i/{image_name}/thumbnail",
     operation_id="get_image_thumbnail",
     response_class=Response,
     responses={
@@ -216,7 +217,7 @@ async def get_image_thumbnail(
 
 
 @images_router.get(
-    "/{image_name}/urls",
+    "/i/{image_name}/urls",
     operation_id="get_image_urls",
     response_model=ImageUrlsDTO,
 )
@@ -265,3 +266,24 @@ async def list_image_dtos(
     )
 
     return image_dtos
+
+
+class DeleteImagesFromListResult(BaseModel):
+    deleted_images: list[str]
+
+
+@images_router.post("/delete", operation_id="delete_images_from_list", response_model=DeleteImagesFromListResult)
+async def delete_images_from_list(
+    image_names: list[str] = Body(description="The list of names of images to delete", embed=True),
+) -> DeleteImagesFromListResult:
+    try:
+        deleted_images: list[str] = []
+        for image_name in image_names:
+            try:
+                ApiDependencies.invoker.services.images.delete(image_name)
+                deleted_images.append(image_name)
+            except:
+                pass
+        return DeleteImagesFromListResult(deleted_images=deleted_images)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail="Failed to delete images")

invokeai/app/cli_app.py

@@ -37,6 +37,7 @@ from invokeai.app.services.image_record_storage import SqliteImageRecordStorage
 from invokeai.app.services.images import ImageService, ImageServiceDependencies
 from invokeai.app.services.resource_name import SimpleNameService
 from invokeai.app.services.urls import LocalUrlService
+from invokeai.app.services.invocation_stats import InvocationStatsService
 from .services.default_graphs import default_text_to_image_graph_id, create_system_graphs
 from .services.latent_storage import DiskLatentsStorage, ForwardCacheLatentsStorage
 
@@ -311,6 +312,7 @@ def invoke_cli():
         graph_library=SqliteItemStorage[LibraryGraph](filename=db_location, table_name="graphs"),
         graph_execution_manager=graph_execution_manager,
         processor=DefaultInvocationProcessor(),
+        performance_statistics=InvocationStatsService(graph_execution_manager),
         logger=logger,
         configuration=config,
     )

invokeai/app/invocations/compel.py

@@ -1,6 +1,14 @@
 from typing import Literal, Optional, Union, List, Annotated
 from pydantic import BaseModel, Field
 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, ModelPatcher
+
 import torch
 from compel import Compel, ReturnedEmbeddingsType
 from compel.prompt_parser import Blend, Conjunction, CrossAttentionControlSubstitute, FlattenedPrompt, Fragment
@@ -101,12 +109,15 @@ class CompelInvocation(BaseInvocation):
             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=context,
-                    ).context.model
+                    (
+                        name,
+                        context.services.model_manager.get_model(
+                            model_name=name,
+                            base_model=self.clip.text_encoder.base_model,
+                            model_type=ModelType.TextualInversion,
+                            context=context,
+                        ).context.model,
+                    )
                 )
             except ModelNotFoundException:
                 # print(e)
@@ -165,7 +176,7 @@ class CompelInvocation(BaseInvocation):
 
 
 class SDXLPromptInvocationBase:
-    def run_clip_raw(self, context, clip_field, prompt, get_pooled):
+    def run_clip_raw(self, context, clip_field, prompt, get_pooled, lora_prefix):
         tokenizer_info = context.services.model_manager.get_model(
             **clip_field.tokenizer.dict(),
             context=context,
@@ -189,12 +200,15 @@ class SDXLPromptInvocationBase:
             name = trigger[1:-1]
             try:
                 ti_list.append(
-                    context.services.model_manager.get_model(
-                        model_name=name,
-                        base_model=clip_field.text_encoder.base_model,
-                        model_type=ModelType.TextualInversion,
-                        context=context,
-                    ).context.model
+                    (
+                        name,
+                        context.services.model_manager.get_model(
+                            model_name=name,
+                            base_model=clip_field.text_encoder.base_model,
+                            model_type=ModelType.TextualInversion,
+                            context=context,
+                        ).context.model,
+                    )
                 )
             except ModelNotFoundException:
                 # print(e)
@@ -202,8 +216,8 @@ class SDXLPromptInvocationBase:
                 # print(traceback.format_exc())
                 print(f'Warn: trigger: "{trigger}" not found')
 
-        with ModelPatcher.apply_lora_text_encoder(
-            text_encoder_info.context.model, _lora_loader()
+        with ModelPatcher.apply_lora(
+            text_encoder_info.context.model, _lora_loader(), lora_prefix
         ), ModelPatcher.apply_ti(tokenizer_info.context.model, text_encoder_info.context.model, ti_list) as (
             tokenizer,
             ti_manager,
@@ -239,7 +253,7 @@ class SDXLPromptInvocationBase:
 
         return c, c_pooled, None
 
-    def run_clip_compel(self, context, clip_field, prompt, get_pooled):
+    def run_clip_compel(self, context, clip_field, prompt, get_pooled, lora_prefix):
         tokenizer_info = context.services.model_manager.get_model(
             **clip_field.tokenizer.dict(),
             context=context,
@@ -263,12 +277,15 @@ class SDXLPromptInvocationBase:
             name = trigger[1:-1]
             try:
                 ti_list.append(
-                    context.services.model_manager.get_model(
-                        model_name=name,
-                        base_model=clip_field.text_encoder.base_model,
-                        model_type=ModelType.TextualInversion,
-                        context=context,
-                    ).context.model
+                    (
+                        name,
+                        context.services.model_manager.get_model(
+                            model_name=name,
+                            base_model=clip_field.text_encoder.base_model,
+                            model_type=ModelType.TextualInversion,
+                            context=context,
+                        ).context.model,
+                    )
                 )
             except ModelNotFoundException:
                 # print(e)
@@ -276,8 +293,8 @@ class SDXLPromptInvocationBase:
                 # print(traceback.format_exc())
                 print(f'Warn: trigger: "{trigger}" not found')
 
-        with ModelPatcher.apply_lora_text_encoder(
-            text_encoder_info.context.model, _lora_loader()
+        with ModelPatcher.apply_lora(
+            text_encoder_info.context.model, _lora_loader(), lora_prefix
         ), ModelPatcher.apply_ti(tokenizer_info.context.model, text_encoder_info.context.model, ti_list) as (
             tokenizer,
             ti_manager,
@@ -349,11 +366,11 @@ class SDXLCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> CompelOutput:
-        c1, c1_pooled, ec1 = self.run_clip_compel(context, self.clip, self.prompt, False)
+        c1, c1_pooled, ec1 = self.run_clip_compel(context, self.clip, self.prompt, False, "lora_te1_")
         if self.style.strip() == "":
-            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.prompt, True)
+            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.prompt, True, "lora_te2_")
         else:
-            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True)
+            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True, "lora_te2_")
 
         original_size = (self.original_height, self.original_width)
         crop_coords = (self.crop_top, self.crop_left)
@@ -407,7 +424,8 @@ class SDXLRefinerCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> CompelOutput:
-        c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True)
+        # TODO: if there will appear lora for refiner - write proper prefix
+        c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True, "<NONE>")
 
         original_size = (self.original_height, self.original_width)
         crop_coords = (self.crop_top, self.crop_left)
@@ -459,11 +477,11 @@ class SDXLRawPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> CompelOutput:
-        c1, c1_pooled, ec1 = self.run_clip_raw(context, self.clip, self.prompt, False)
+        c1, c1_pooled, ec1 = self.run_clip_raw(context, self.clip, self.prompt, False, "lora_te1_")
         if self.style.strip() == "":
-            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.prompt, True)
+            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.prompt, True, "lora_te2_")
         else:
-            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True)
+            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True, "lora_te2_")
 
         original_size = (self.original_height, self.original_width)
         crop_coords = (self.crop_top, self.crop_left)
@@ -517,7 +535,8 @@ class SDXLRefinerRawPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> CompelOutput:
-        c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True)
+        # TODO: if there will appear lora for refiner - write proper prefix
+        c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True, "<NONE>")
 
         original_size = (self.original_height, self.original_width)
         crop_coords = (self.crop_top, self.crop_left)

invokeai/app/invocations/image.py

@@ -3,6 +3,7 @@
 from typing import Literal, Optional
 
 import numpy
+import cv2
 from PIL import Image, ImageFilter, ImageOps, ImageChops
 from pydantic import Field
 from pathlib import Path
@@ -650,3 +651,143 @@ class ImageWatermarkInvocation(BaseInvocation, PILInvocationConfig):
             width=image_dto.width,
             height=image_dto.height,
         )
+
+
+class ImageHueAdjustmentInvocation(BaseInvocation):
+    """Adjusts the Hue of an image."""
+
+    # fmt: off
+    type: Literal["img_hue_adjust"] = "img_hue_adjust"
+
+    # Inputs
+    image: ImageField = Field(default=None, description="The image to adjust")
+    hue: int = Field(default=0, description="The degrees by which to rotate the hue, 0-360")
+    # fmt: on
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        pil_image = context.services.images.get_pil_image(self.image.image_name)
+
+        # Convert image to HSV color space
+        hsv_image = numpy.array(pil_image.convert("HSV"))
+
+        # Convert hue from 0..360 to 0..256
+        hue = int(256 * ((self.hue % 360) / 360))
+
+        # Increment each hue and wrap around at 255
+        hsv_image[:, :, 0] = (hsv_image[:, :, 0] + hue) % 256
+
+        # Convert back to PIL format and to original color mode
+        pil_image = Image.fromarray(hsv_image, mode="HSV").convert("RGBA")
+
+        image_dto = context.services.images.create(
+            image=pil_image,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            is_intermediate=self.is_intermediate,
+            session_id=context.graph_execution_state_id,
+        )
+
+        return ImageOutput(
+            image=ImageField(
+                image_name=image_dto.image_name,
+            ),
+            width=image_dto.width,
+            height=image_dto.height,
+        )
+
+
+class ImageLuminosityAdjustmentInvocation(BaseInvocation):
+    """Adjusts the Luminosity (Value) of an image."""
+
+    # fmt: off
+    type: Literal["img_luminosity_adjust"] = "img_luminosity_adjust"
+
+    # Inputs
+    image: ImageField = Field(default=None, description="The image to adjust")
+    luminosity: float = Field(default=1.0, ge=0, le=1, description="The factor by which to adjust the luminosity (value)")
+    # fmt: on
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        pil_image = context.services.images.get_pil_image(self.image.image_name)
+
+        # Convert PIL image to OpenCV format (numpy array), note color channel
+        # ordering is changed from RGB to BGR
+        image = numpy.array(pil_image.convert("RGB"))[:, :, ::-1]
+
+        # Convert image to HSV color space
+        hsv_image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
+
+        # Adjust the luminosity (value)
+        hsv_image[:, :, 2] = numpy.clip(hsv_image[:, :, 2] * self.luminosity, 0, 255)
+
+        # Convert image back to BGR color space
+        image = cv2.cvtColor(hsv_image, cv2.COLOR_HSV2BGR)
+
+        # Convert back to PIL format and to original color mode
+        pil_image = Image.fromarray(image[:, :, ::-1], "RGB").convert("RGBA")
+
+        image_dto = context.services.images.create(
+            image=pil_image,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            is_intermediate=self.is_intermediate,
+            session_id=context.graph_execution_state_id,
+        )
+
+        return ImageOutput(
+            image=ImageField(
+                image_name=image_dto.image_name,
+            ),
+            width=image_dto.width,
+            height=image_dto.height,
+        )
+
+
+class ImageSaturationAdjustmentInvocation(BaseInvocation):
+    """Adjusts the Saturation of an image."""
+
+    # fmt: off
+    type: Literal["img_saturation_adjust"] = "img_saturation_adjust"
+
+    # Inputs
+    image: ImageField = Field(default=None, description="The image to adjust")
+    saturation: float = Field(default=1.0, ge=0, le=1, description="The factor by which to adjust the saturation")
+    # fmt: on
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        pil_image = context.services.images.get_pil_image(self.image.image_name)
+
+        # Convert PIL image to OpenCV format (numpy array), note color channel
+        # ordering is changed from RGB to BGR
+        image = numpy.array(pil_image.convert("RGB"))[:, :, ::-1]
+
+        # Convert image to HSV color space
+        hsv_image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
+
+        # Adjust the saturation
+        hsv_image[:, :, 1] = numpy.clip(hsv_image[:, :, 1] * self.saturation, 0, 255)
+
+        # Convert image back to BGR color space
+        image = cv2.cvtColor(hsv_image, cv2.COLOR_HSV2BGR)
+
+        # Convert back to PIL format and to original color mode
+        pil_image = Image.fromarray(image[:, :, ::-1], "RGB").convert("RGBA")
+
+        image_dto = context.services.images.create(
+            image=pil_image,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            is_intermediate=self.is_intermediate,
+            session_id=context.graph_execution_state_id,
+        )
+
+        return ImageOutput(
+            image=ImageField(
+                image_name=image_dto.image_name,
+            ),
+            width=image_dto.width,
+            height=image_dto.height,
+        )

invokeai/app/invocations/latent.py

@@ -14,7 +14,7 @@ from invokeai.app.invocations.metadata import CoreMetadata
 from invokeai.app.util.step_callback import stable_diffusion_step_callback
 from invokeai.backend.model_management.models import ModelType, SilenceWarnings
 
-from ...backend.model_management.lora import ModelPatcher
+from ...backend.model_management import ModelPatcher
 from ...backend.stable_diffusion import PipelineIntermediateState
 from ...backend.stable_diffusion.diffusers_pipeline import (
     ConditioningData,
@@ -24,6 +24,7 @@ from ...backend.stable_diffusion.diffusers_pipeline import (
 )
 from ...backend.stable_diffusion.diffusion.shared_invokeai_diffusion import PostprocessingSettings
 from ...backend.stable_diffusion.schedulers import SCHEDULER_MAP
+from ...backend.model_management import ModelPatcher
 from ...backend.util.devices import choose_torch_device, torch_dtype, choose_precision
 from ..models.image import ImageCategory, ImageField, ResourceOrigin
 from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext

invokeai/app/invocations/metadata.py

@@ -1,6 +1,6 @@
 from typing import Literal, Optional, Union
 
-from pydantic import BaseModel, Field
+from pydantic import Field
 
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
@@ -10,16 +10,17 @@ from invokeai.app.invocations.baseinvocation import (
 )
 from invokeai.app.invocations.controlnet_image_processors import ControlField
 from invokeai.app.invocations.model import LoRAModelField, MainModelField, VAEModelField
+from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
 
 
-class LoRAMetadataField(BaseModel):
+class LoRAMetadataField(BaseModelExcludeNull):
     """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):
+class CoreMetadata(BaseModelExcludeNull):
     """Core generation metadata for an image generated in InvokeAI."""
 
     generation_mode: str = Field(
@@ -70,7 +71,7 @@ class CoreMetadata(BaseModel):
     refiner_start: Union[float, None] = Field(default=None, description="The start value used for refiner denoising")
 
 
-class ImageMetadata(BaseModel):
+class ImageMetadata(BaseModelExcludeNull):
     """An image's generation metadata"""
 
     metadata: Optional[dict] = Field(

invokeai/app/invocations/model.py

@@ -53,6 +53,7 @@ class MainModelField(BaseModel):
 
     model_name: str = Field(description="Name of the model")
     base_model: BaseModelType = Field(description="Base model")
+    model_type: ModelType = Field(description="Model Type")
 
 
 class LoRAModelField(BaseModel):
@@ -261,6 +262,103 @@ class LoraLoaderInvocation(BaseInvocation):
         return output
 
 
+class SDXLLoraLoaderOutput(BaseInvocationOutput):
+    """Model loader output"""
+
+    # fmt: off
+    type: Literal["sdxl_lora_loader_output"] = "sdxl_lora_loader_output"
+
+    unet: Optional[UNetField] = Field(default=None, description="UNet submodel")
+    clip: Optional[ClipField] = Field(default=None, description="Tokenizer and text_encoder submodels")
+    clip2: Optional[ClipField] = Field(default=None, description="Tokenizer2 and text_encoder2 submodels")
+    # fmt: on
+
+
+class SDXLLoraLoaderInvocation(BaseInvocation):
+    """Apply selected lora to unet and text_encoder."""
+
+    type: Literal["sdxl_lora_loader"] = "sdxl_lora_loader"
+
+    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")
+    clip2: Optional[ClipField] = Field(description="Clip2 model for applying lora")
+
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Lora Loader",
+                "tags": ["lora", "loader"],
+                "type_hints": {"lora": "lora_model"},
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> SDXLLoraLoaderOutput:
+        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=lora_name,
+            model_type=ModelType.Lora,
+        ):
+            raise Exception(f"Unknown lora name: {lora_name}!")
+
+        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 == lora_name for lora in self.clip.loras):
+            raise Exception(f'Lora "{lora_name}" already applied to clip')
+
+        if self.clip2 is not None and any(lora.model_name == lora_name for lora in self.clip2.loras):
+            raise Exception(f'Lora "{lora_name}" already applied to clip2')
+
+        output = SDXLLoraLoaderOutput()
+
+        if self.unet is not None:
+            output.unet = copy.deepcopy(self.unet)
+            output.unet.loras.append(
+                LoraInfo(
+                    base_model=base_model,
+                    model_name=lora_name,
+                    model_type=ModelType.Lora,
+                    submodel=None,
+                    weight=self.weight,
+                )
+            )
+
+        if self.clip is not None:
+            output.clip = copy.deepcopy(self.clip)
+            output.clip.loras.append(
+                LoraInfo(
+                    base_model=base_model,
+                    model_name=lora_name,
+                    model_type=ModelType.Lora,
+                    submodel=None,
+                    weight=self.weight,
+                )
+            )
+
+        if self.clip2 is not None:
+            output.clip2 = copy.deepcopy(self.clip2)
+            output.clip2.loras.append(
+                LoraInfo(
+                    base_model=base_model,
+                    model_name=lora_name,
+                    model_type=ModelType.Lora,
+                    submodel=None,
+                    weight=self.weight,
+                )
+            )
+
+        return output
+
+
 class VAEModelField(BaseModel):
     """Vae model field"""
 

invokeai/app/invocations/onnx.py

@@ -0,0 +1,573 @@
+# Copyright (c) 2023 Borisov Sergey (https://github.com/StAlKeR7779)
+
+from contextlib import ExitStack
+from typing import List, Literal, Optional, Union
+
+import re
+import inspect
+
+from pydantic import BaseModel, Field, validator
+import torch
+import numpy as np
+from diffusers import ControlNetModel, DPMSolverMultistepScheduler
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.schedulers import SchedulerMixin as Scheduler
+
+from ..models.image import ImageCategory, ImageField, ResourceOrigin
+from ...backend.model_management import ONNXModelPatcher
+from ...backend.util import choose_torch_device
+from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
+from .compel import ConditioningField
+from .controlnet_image_processors import ControlField
+from .image import ImageOutput
+from .model import ModelInfo, UNetField, VaeField
+
+from invokeai.app.invocations.metadata import CoreMetadata
+from invokeai.backend import BaseModelType, ModelType, SubModelType
+from invokeai.app.util.step_callback import stable_diffusion_step_callback
+from ...backend.stable_diffusion import PipelineIntermediateState
+
+from tqdm import tqdm
+from .model import ClipField
+from .latent import LatentsField, LatentsOutput, build_latents_output, get_scheduler, SAMPLER_NAME_VALUES
+from .compel import CompelOutput
+
+
+ORT_TO_NP_TYPE = {
+    "tensor(bool)": np.bool_,
+    "tensor(int8)": np.int8,
+    "tensor(uint8)": np.uint8,
+    "tensor(int16)": np.int16,
+    "tensor(uint16)": np.uint16,
+    "tensor(int32)": np.int32,
+    "tensor(uint32)": np.uint32,
+    "tensor(int64)": np.int64,
+    "tensor(uint64)": np.uint64,
+    "tensor(float16)": np.float16,
+    "tensor(float)": np.float32,
+    "tensor(double)": np.float64,
+}
+
+PRECISION_VALUES = Literal[tuple(list(ORT_TO_NP_TYPE.keys()))]
+
+
+class ONNXPromptInvocation(BaseInvocation):
+    type: Literal["prompt_onnx"] = "prompt_onnx"
+
+    prompt: str = Field(default="", description="Prompt")
+    clip: ClipField = Field(None, description="Clip to use")
+
+    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, ExitStack() as stack:
+            loras = [
+                (context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight)
+                for lora in self.clip.loras
+            ]
+
+            ti_list = []
+            for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", self.prompt):
+                name = trigger[1:-1]
+                try:
+                    ti_list.append(
+                        (
+                            name,
+                            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')
+            if loras or ti_list:
+                text_encoder.release_session()
+            with ONNXModelPatcher.apply_lora_text_encoder(text_encoder, loras), ONNXModelPatcher.apply_ti(
+                orig_tokenizer, text_encoder, ti_list
+            ) as (tokenizer, ti_manager):
+                text_encoder.create_session()
+
+                # copy from
+                # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L153
+                text_inputs = tokenizer(
+                    self.prompt,
+                    padding="max_length",
+                    max_length=tokenizer.model_max_length,
+                    truncation=True,
+                    return_tensors="np",
+                )
+                text_input_ids = text_inputs.input_ids
+                """
+                untruncated_ids = tokenizer(prompt, padding="max_length", return_tensors="np").input_ids
+
+                if not np.array_equal(text_input_ids, untruncated_ids):
+                    removed_text = self.tokenizer.batch_decode(
+                        untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+                    )
+                    logger.warning(
+                        "The following part of your input was truncated because CLIP can only handle sequences up to"
+                        f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+                    )
+                """
+
+                prompt_embeds = text_encoder(input_ids=text_input_ids.astype(np.int32))[0]
+
+        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, (prompt_embeds, None))
+
+        return CompelOutput(
+            conditioning=ConditioningField(
+                conditioning_name=conditioning_name,
+            ),
+        )
+
+
+# Text to image
+class ONNXTextToLatentsInvocation(BaseInvocation):
+    """Generates latents from conditionings."""
+
+    type: Literal["t2l_onnx"] = "t2l_onnx"
+
+    # Inputs
+    # fmt: off
+    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
+    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
+    noise: Optional[LatentsField] = Field(description="The noise to use")
+    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
+    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
+    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
+    precision: PRECISION_VALUES = Field(default = "tensor(float16)", description="The precision to use when generating latents")
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
+    # seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
+    # seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
+    # fmt: on
+
+    @validator("cfg_scale")
+    def ge_one(cls, v):
+        """validate that all cfg_scale values are >= 1"""
+        if isinstance(v, list):
+            for i in v:
+                if i < 1:
+                    raise ValueError("cfg_scale must be greater than 1")
+        else:
+            if v < 1:
+                raise ValueError("cfg_scale must be greater than 1")
+        return v
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "tags": ["latents"],
+                "type_hints": {
+                    "model": "model",
+                    "control": "control",
+                    # "cfg_scale": "float",
+                    "cfg_scale": "number",
+                },
+            },
+        }
+
+    # based on
+    # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
+    def invoke(self, context: InvocationContext) -> LatentsOutput:
+        c, _ = context.services.latents.get(self.positive_conditioning.conditioning_name)
+        uc, _ = context.services.latents.get(self.negative_conditioning.conditioning_name)
+        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]
+        if isinstance(c, torch.Tensor):
+            c = c.cpu().numpy()
+        if isinstance(uc, torch.Tensor):
+            uc = uc.cpu().numpy()
+        device = torch.device(choose_torch_device())
+        prompt_embeds = np.concatenate([uc, c])
+
+        latents = context.services.latents.get(self.noise.latents_name)
+        if isinstance(latents, torch.Tensor):
+            latents = latents.cpu().numpy()
+
+        # TODO: better execution device handling
+        latents = latents.astype(ORT_TO_NP_TYPE[self.precision])
+
+        # get the initial random noise unless the user supplied it
+        do_classifier_free_guidance = True
+        # latents_dtype = prompt_embeds.dtype
+        # latents_shape = (batch_size * num_images_per_prompt, 4, height // 8, width // 8)
+        # if latents.shape != latents_shape:
+        #    raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
+
+        scheduler = get_scheduler(
+            context=context,
+            scheduler_info=self.unet.scheduler,
+            scheduler_name=self.scheduler,
+        )
+
+        def torch2numpy(latent: torch.Tensor):
+            return latent.cpu().numpy()
+
+        def numpy2torch(latent, device):
+            return torch.from_numpy(latent).to(device)
+
+        def dispatch_progress(
+            self, context: InvocationContext, source_node_id: str, intermediate_state: PipelineIntermediateState
+        ) -> None:
+            stable_diffusion_step_callback(
+                context=context,
+                intermediate_state=intermediate_state,
+                node=self.dict(),
+                source_node_id=source_node_id,
+            )
+
+        scheduler.set_timesteps(self.steps)
+        latents = latents * np.float64(scheduler.init_noise_sigma)
+
+        extra_step_kwargs = dict()
+        if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
+            extra_step_kwargs.update(
+                eta=0.0,
+            )
+
+        unet_info = context.services.model_manager.get_model(**self.unet.unet.dict())
+
+        with unet_info as unet, 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]
+            loras = [
+                (context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight)
+                for lora in self.unet.loras
+            ]
+
+            if loras:
+                unet.release_session()
+            with ONNXModelPatcher.apply_lora_unet(unet, loras):
+                # TODO:
+                _, _, h, w = latents.shape
+                unet.create_session(h, w)
+
+                timestep_dtype = next(
+                    (input.type for input in unet.session.get_inputs() if input.name == "timestep"), "tensor(float16)"
+                )
+                timestep_dtype = ORT_TO_NP_TYPE[timestep_dtype]
+                for i in tqdm(range(len(scheduler.timesteps))):
+                    t = scheduler.timesteps[i]
+                    # expand the latents if we are doing classifier free guidance
+                    latent_model_input = np.concatenate([latents] * 2) if do_classifier_free_guidance else latents
+                    latent_model_input = scheduler.scale_model_input(numpy2torch(latent_model_input, device), t)
+                    latent_model_input = latent_model_input.cpu().numpy()
+
+                    # predict the noise residual
+                    timestep = np.array([t], dtype=timestep_dtype)
+                    noise_pred = unet(sample=latent_model_input, timestep=timestep, encoder_hidden_states=prompt_embeds)
+                    noise_pred = noise_pred[0]
+
+                    # perform guidance
+                    if do_classifier_free_guidance:
+                        noise_pred_uncond, noise_pred_text = np.split(noise_pred, 2)
+                        noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
+
+                    # compute the previous noisy sample x_t -> x_t-1
+                    scheduler_output = scheduler.step(
+                        numpy2torch(noise_pred, device), t, numpy2torch(latents, device), **extra_step_kwargs
+                    )
+                    latents = torch2numpy(scheduler_output.prev_sample)
+
+                    state = PipelineIntermediateState(
+                        run_id="test", step=i, timestep=timestep, latents=scheduler_output.prev_sample
+                    )
+                    dispatch_progress(self, context=context, source_node_id=source_node_id, intermediate_state=state)
+
+                    # call the callback, if provided
+                    # if callback is not None and i % callback_steps == 0:
+                    #    callback(i, t, latents)
+
+        torch.cuda.empty_cache()
+
+        name = f"{context.graph_execution_state_id}__{self.id}"
+        context.services.latents.save(name, latents)
+        return build_latents_output(latents_name=name, latents=torch.from_numpy(latents))
+
+
+# Latent to image
+class ONNXLatentsToImageInvocation(BaseInvocation):
+    """Generates an image from latents."""
+
+    type: Literal["l2i_onnx"] = "l2i_onnx"
+
+    # Inputs
+    latents: Optional[LatentsField] = Field(description="The latents to generate an image from")
+    vae: VaeField = Field(default=None, description="Vae submodel")
+    metadata: Optional[CoreMetadata] = Field(
+        default=None, description="Optional core metadata to be written to the image"
+    )
+    # tiled: bool = Field(default=False, description="Decode latents by overlaping tiles(less memory consumption)")
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "tags": ["latents", "image"],
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        latents = context.services.latents.get(self.latents.latents_name)
+
+        if self.vae.vae.submodel != SubModelType.VaeDecoder:
+            raise Exception(f"Expected vae_decoder, found: {self.vae.vae.model_type}")
+
+        vae_info = context.services.model_manager.get_model(
+            **self.vae.vae.dict(),
+        )
+
+        # clear memory as vae decode can request a lot
+        torch.cuda.empty_cache()
+
+        with vae_info as vae:
+            vae.create_session()
+
+            # copied from
+            # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L427
+            latents = 1 / 0.18215 * latents
+            # image = self.vae_decoder(latent_sample=latents)[0]
+            # it seems likes there is a strange result for using half-precision vae decoder if batchsize>1
+            image = np.concatenate([vae(latent_sample=latents[i : i + 1])[0] for i in range(latents.shape[0])])
+
+            image = np.clip(image / 2 + 0.5, 0, 1)
+            image = image.transpose((0, 2, 3, 1))
+            image = VaeImageProcessor.numpy_to_pil(image)[0]
+
+        torch.cuda.empty_cache()
+
+        image_dto = context.services.images.create(
+            image=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,
+            metadata=self.metadata.dict() if self.metadata else None,
+        )
+
+        return ImageOutput(
+            image=ImageField(image_name=image_dto.image_name),
+            width=image_dto.width,
+            height=image_dto.height,
+        )
+
+
+class ONNXModelLoaderOutput(BaseInvocationOutput):
+    """Model loader output"""
+
+    # fmt: off
+    type: Literal["model_loader_output_onnx"] = "model_loader_output_onnx"
+
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    clip: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
+    vae_decoder: VaeField = Field(default=None, description="Vae submodel")
+    vae_encoder: VaeField = Field(default=None, description="Vae submodel")
+    # fmt: on
+
+
+class ONNXSD1ModelLoaderInvocation(BaseInvocation):
+    """Loading submodels of selected model."""
+
+    type: Literal["sd1_model_loader_onnx"] = "sd1_model_loader_onnx"
+
+    model_name: str = Field(default="", description="Model to load")
+    # TODO: precision?
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {"tags": ["model", "loader"], "type_hints": {"model_name": "model"}},  # TODO: rename to model_name?
+        }
+
+    def invoke(self, context: InvocationContext) -> ONNXModelLoaderOutput:
+        model_name = "stable-diffusion-v1-5"
+        base_model = BaseModelType.StableDiffusion1
+
+        # TODO: not found exceptions
+        if not context.services.model_manager.model_exists(
+            model_name=model_name,
+            base_model=BaseModelType.StableDiffusion1,
+            model_type=ModelType.ONNX,
+        ):
+            raise Exception(f"Unkown model name: {model_name}!")
+
+        return ONNXModelLoaderOutput(
+            unet=UNetField(
+                unet=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.UNet,
+                ),
+                scheduler=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.Scheduler,
+                ),
+                loras=[],
+            ),
+            clip=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.Tokenizer,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.TextEncoder,
+                ),
+                loras=[],
+            ),
+            vae_decoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.VaeDecoder,
+                ),
+            ),
+            vae_encoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.VaeEncoder,
+                ),
+            ),
+        )
+
+
+class OnnxModelField(BaseModel):
+    """Onnx model field"""
+
+    model_name: str = Field(description="Name of the model")
+    base_model: BaseModelType = Field(description="Base model")
+    model_type: ModelType = Field(description="Model Type")
+
+
+class OnnxModelLoaderInvocation(BaseInvocation):
+    """Loads a main model, outputting its submodels."""
+
+    type: Literal["onnx_model_loader"] = "onnx_model_loader"
+
+    model: OnnxModelField = Field(description="The model to load")
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Onnx Model Loader",
+                "tags": ["model", "loader"],
+                "type_hints": {"model": "model"},
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> ONNXModelLoaderOutput:
+        base_model = self.model.base_model
+        model_name = self.model.model_name
+        model_type = ModelType.ONNX
+
+        # TODO: not found exceptions
+        if not context.services.model_manager.model_exists(
+            model_name=model_name,
+            base_model=base_model,
+            model_type=model_type,
+        ):
+            raise Exception(f"Unknown {base_model} {model_type} model: {model_name}")
+
+        """
+        if not context.services.model_manager.model_exists(
+            model_name=self.model_name,
+            model_type=SDModelType.Diffusers,
+            submodel=SDModelType.Tokenizer,
+        ):
+            raise Exception(
+                f"Failed to find tokenizer submodel in {self.model_name}! Check if model corrupted"
+            )
+
+        if not context.services.model_manager.model_exists(
+            model_name=self.model_name,
+            model_type=SDModelType.Diffusers,
+            submodel=SDModelType.TextEncoder,
+        ):
+            raise Exception(
+                f"Failed to find text_encoder submodel in {self.model_name}! Check if model corrupted"
+            )
+
+        if not context.services.model_manager.model_exists(
+            model_name=self.model_name,
+            model_type=SDModelType.Diffusers,
+            submodel=SDModelType.UNet,
+        ):
+            raise Exception(
+                f"Failed to find unet submodel from {self.model_name}! Check if model corrupted"
+            )
+        """
+
+        return ONNXModelLoaderOutput(
+            unet=UNetField(
+                unet=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.UNet,
+                ),
+                scheduler=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Scheduler,
+                ),
+                loras=[],
+            ),
+            clip=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Tokenizer,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.TextEncoder,
+                ),
+                loras=[],
+                skipped_layers=0,
+            ),
+            vae_decoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.VaeDecoder,
+                ),
+            ),
+            vae_encoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.VaeEncoder,
+                ),
+            ),
+        )

invokeai/app/invocations/params.py

@@ -4,6 +4,8 @@ from typing import Literal
 
 from pydantic import Field
 
+from invokeai.app.invocations.prompt import PromptOutput
+
 from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
 from .math import FloatOutput, IntOutput
 
@@ -64,3 +66,18 @@ class ParamStringInvocation(BaseInvocation):
 
     def invoke(self, context: InvocationContext) -> StringOutput:
         return StringOutput(text=self.text)
+
+
+class ParamPromptInvocation(BaseInvocation):
+    """A prompt input parameter"""
+
+    type: Literal["param_prompt"] = "param_prompt"
+    prompt: str = Field(default="", description="The prompt value")
+
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {"tags": ["param", "prompt"], "title": "Prompt"},
+        }
+
+    def invoke(self, context: InvocationContext) -> PromptOutput:
+        return PromptOutput(prompt=self.prompt)

invokeai/app/invocations/sdxl.py

@@ -5,7 +5,7 @@ from typing import List, Literal, Optional, Union
 
 from pydantic import Field, validator
 
-from ...backend.model_management import ModelType, SubModelType
+from ...backend.model_management import ModelType, SubModelType, ModelPatcher
 from invokeai.app.util.step_callback import stable_diffusion_xl_step_callback
 from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
 
@@ -293,10 +293,20 @@ class SDXLTextToLatentsInvocation(BaseInvocation):
 
         num_inference_steps = self.steps
 
+        def _lora_loader():
+            for lora in self.unet.loras:
+                lora_info = context.services.model_manager.get_model(
+                    **lora.dict(exclude={"weight"}),
+                    context=context,
+                )
+                yield (lora_info.context.model, lora.weight)
+                del lora_info
+            return
+
         unet_info = context.services.model_manager.get_model(**self.unet.unet.dict(), context=context)
         do_classifier_free_guidance = True
         cross_attention_kwargs = None
-        with unet_info as unet:
+        with ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()), unet_info as unet:
             scheduler.set_timesteps(num_inference_steps, device=unet.device)
             timesteps = scheduler.timesteps
 
@@ -543,9 +553,19 @@ class SDXLLatentsToLatentsInvocation(BaseInvocation):
             context=context,
         )
 
+        def _lora_loader():
+            for lora in self.unet.loras:
+                lora_info = context.services.model_manager.get_model(
+                    **lora.dict(exclude={"weight"}),
+                    context=context,
+                )
+                yield (lora_info.context.model, lora.weight)
+                del lora_info
+            return
+
         do_classifier_free_guidance = True
         cross_attention_kwargs = None
-        with unet_info as unet:
+        with ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()), unet_info as unet:
             # apply denoising_start
             num_inference_steps = self.steps
             scheduler.set_timesteps(num_inference_steps, device=unet.device)

invokeai/app/services/board_image_record_storage.py

@@ -25,7 +25,6 @@ class BoardImageRecordStorageBase(ABC):
     @abstractmethod
     def remove_image_from_board(
         self,
-        board_id: str,
         image_name: str,
     ) -> None:
         """Removes an image from a board."""
@@ -154,7 +153,6 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
 
     def remove_image_from_board(
         self,
-        board_id: str,
         image_name: str,
     ) -> None:
         try:
@@ -162,9 +160,9 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
             self._cursor.execute(
                 """--sql
                 DELETE FROM board_images
-                WHERE board_id = ? AND image_name = ?;
+                WHERE image_name = ?;
                 """,
-                (board_id, image_name),
+                (image_name,),
             )
             self._conn.commit()
         except sqlite3.Error as e:

invokeai/app/services/board_images.py

@@ -31,7 +31,6 @@ class BoardImagesServiceABC(ABC):
     @abstractmethod
     def remove_image_from_board(
         self,
-        board_id: str,
         image_name: str,
     ) -> None:
         """Removes an image from a board."""
@@ -93,10 +92,9 @@ class BoardImagesService(BoardImagesServiceABC):
 
     def remove_image_from_board(
         self,
-        board_id: str,
         image_name: str,
     ) -> None:
-        self._services.board_image_records.remove_image_from_board(board_id, image_name)
+        self._services.board_image_records.remove_image_from_board(image_name)
 
     def get_all_board_image_names_for_board(
         self,

invokeai/app/services/config.py

@@ -171,7 +171,6 @@ from pydantic import BaseSettings, Field, parse_obj_as
 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")
 
@@ -356,8 +355,8 @@ class InvokeAISettings(BaseSettings):
 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 any([(venv.parent / x).exists() for x in [INIT_FILE, LEGACY_INIT_FILE, MODEL_CORE]]):
+        root = Path(os.environ["INVOKEAI_ROOT"])
+    elif any([(venv.parent / x).exists() for x in [INIT_FILE, LEGACY_INIT_FILE]]):
         root = (venv.parent).resolve()
     else:
         root = Path("~/invokeai").expanduser().resolve()
@@ -403,7 +402,7 @@ class InvokeAIAppConfig(InvokeAISettings):
     xformers_enabled    : bool = Field(default=True, description="Enable/disable memory-efficient attention", category='Memory/Performance')
     tiled_decode        : bool = Field(default=False, description="Whether to enable tiled VAE decode (reduces memory consumption with some performance penalty)", category='Memory/Performance')
 
-    root                : Path = Field(default=_find_root(), description='InvokeAI runtime root directory', category='Paths')
+    root                : Path = Field(default=None, description='InvokeAI runtime root directory', category='Paths')
     autoimport_dir      : Path = Field(default='autoimport', description='Path to a directory of models files to be imported on startup.', category='Paths')
     lora_dir            : Path = Field(default=None, description='Path to a directory of LoRA/LyCORIS models to be imported on startup.', category='Paths')
     embedding_dir       : Path = Field(default=None, description='Path to a directory of Textual Inversion embeddings to be imported on startup.', category='Paths')
@@ -415,6 +414,7 @@ class InvokeAIAppConfig(InvokeAISettings):
     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')
+    ignore_missing_core_models : bool = Field(default=False, description='Ignore missing models in models/core/convert')
 
     model               : str = Field(default='stable-diffusion-1.5', description='Initial model name', category='Models')
 
@@ -472,9 +472,11 @@ class InvokeAIAppConfig(InvokeAISettings):
         Path to the runtime root directory
         """
         if self.root:
-            return Path(self.root).expanduser().absolute()
+            root = Path(self.root).expanduser().absolute()
         else:
-            return self.find_root()
+            root = self.find_root().expanduser().absolute()
+        self.root = root  # insulate ourselves from relative paths that may change
+        return root
 
     @property
     def root_dir(self) -> Path:

invokeai/app/services/images.py

@@ -289,9 +289,10 @@ class ImageService(ImageServiceABC):
     def get_metadata(self, image_name: str) -> Optional[ImageMetadata]:
         try:
             image_record = self._services.image_records.get(image_name)
+            metadata = self._services.image_records.get_metadata(image_name)
 
             if not image_record.session_id:
-                return ImageMetadata()
+                return ImageMetadata(metadata=metadata)
 
             session_raw = self._services.graph_execution_manager.get_raw(image_record.session_id)
             graph = None
@@ -303,7 +304,6 @@ class ImageService(ImageServiceABC):
                     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")

invokeai/app/services/invocation_services.py

@@ -32,6 +32,7 @@ class InvocationServices:
     logger: "Logger"
     model_manager: "ModelManagerServiceBase"
     processor: "InvocationProcessorABC"
+    performance_statistics: "InvocationStatsServiceBase"
     queue: "InvocationQueueABC"
 
     def __init__(
@@ -47,6 +48,7 @@ class InvocationServices:
         logger: "Logger",
         model_manager: "ModelManagerServiceBase",
         processor: "InvocationProcessorABC",
+        performance_statistics: "InvocationStatsServiceBase",
         queue: "InvocationQueueABC",
     ):
         self.board_images = board_images
@@ -61,4 +63,5 @@ class InvocationServices:
         self.logger = logger
         self.model_manager = model_manager
         self.processor = processor
+        self.performance_statistics = performance_statistics
         self.queue = queue

invokeai/app/services/invocation_stats.py

@@ -0,0 +1,223 @@
+# Copyright 2023 Lincoln D. Stein <lincoln.stein@gmail.com>
+"""Utility to collect execution time and GPU usage stats on invocations in flight"""
+
+"""
+Usage:
+
+statistics = InvocationStatsService(graph_execution_manager)
+with statistics.collect_stats(invocation, graph_execution_state.id):
+      ... execute graphs...
+statistics.log_stats()
+
+Typical output:
+[2023-08-02 18:03:04,507]::[InvokeAI]::INFO --> Graph stats: c7764585-9c68-4d9d-a199-55e8186790f3
+[2023-08-02 18:03:04,507]::[InvokeAI]::INFO --> Node                 Calls  Seconds  VRAM Used
+[2023-08-02 18:03:04,507]::[InvokeAI]::INFO --> main_model_loader        1   0.005s     0.01G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> clip_skip                1   0.004s     0.01G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> compel                   2   0.512s     0.26G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> rand_int                 1   0.001s     0.01G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> range_of_size            1   0.001s     0.01G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> iterate                  1   0.001s     0.01G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> metadata_accumulator     1   0.002s     0.01G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> noise                    1   0.002s     0.01G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> t2l                      1   3.541s     1.93G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> l2i                      1   0.679s     0.58G
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> TOTAL GRAPH EXECUTION TIME:  4.749s
+[2023-08-02 18:03:04,508]::[InvokeAI]::INFO --> Current VRAM utilization 0.01G
+
+The abstract base class for this class is InvocationStatsServiceBase. An implementing class which
+writes to the system log is stored in InvocationServices.performance_statistics.
+"""
+
+import time
+from abc import ABC, abstractmethod
+from contextlib import AbstractContextManager
+from dataclasses import dataclass, field
+from typing import Dict
+
+import torch
+
+import invokeai.backend.util.logging as logger
+
+from ..invocations.baseinvocation import BaseInvocation
+from .graph import GraphExecutionState
+from .item_storage import ItemStorageABC
+
+
+class InvocationStatsServiceBase(ABC):
+    "Abstract base class for recording node memory/time performance statistics"
+
+    @abstractmethod
+    def __init__(self, graph_execution_manager: ItemStorageABC["GraphExecutionState"]):
+        """
+        Initialize the InvocationStatsService and reset counters to zero
+        :param graph_execution_manager: Graph execution manager for this session
+        """
+        pass
+
+    @abstractmethod
+    def collect_stats(
+        self,
+        invocation: BaseInvocation,
+        graph_execution_state_id: str,
+    ) -> AbstractContextManager:
+        """
+        Return a context object that will capture the statistics on the execution
+        of invocaation. Use with: to place around the part of the code that executes the invocation.
+        :param invocation: BaseInvocation object from the current graph.
+        :param graph_execution_state: GraphExecutionState object from the current session.
+        """
+        pass
+
+    @abstractmethod
+    def reset_stats(self, graph_execution_state_id: str):
+        """
+        Reset all statistics for the indicated graph
+        :param graph_execution_state_id
+        """
+        pass
+
+    @abstractmethod
+    def reset_all_stats(self):
+        """Zero all statistics"""
+        pass
+
+    @abstractmethod
+    def update_invocation_stats(
+        self,
+        graph_id: str,
+        invocation_type: str,
+        time_used: float,
+        vram_used: float,
+    ):
+        """
+        Add timing information on execution of a node. Usually
+        used internally.
+        :param graph_id: ID of the graph that is currently executing
+        :param invocation_type: String literal type of the node
+        :param time_used: Time used by node's exection (sec)
+        :param vram_used: Maximum VRAM used during exection (GB)
+        """
+        pass
+
+    @abstractmethod
+    def log_stats(self):
+        """
+        Write out the accumulated statistics to the log or somewhere else.
+        """
+        pass
+
+
+@dataclass
+class NodeStats:
+    """Class for tracking execution stats of an invocation node"""
+
+    calls: int = 0
+    time_used: float = 0.0  # seconds
+    max_vram: float = 0.0  # GB
+
+
+@dataclass
+class NodeLog:
+    """Class for tracking node usage"""
+
+    # {node_type => NodeStats}
+    nodes: Dict[str, NodeStats] = field(default_factory=dict)
+
+
+class InvocationStatsService(InvocationStatsServiceBase):
+    """Accumulate performance information about a running graph. Collects time spent in each node,
+    as well as the maximum and current VRAM utilisation for CUDA systems"""
+
+    def __init__(self, graph_execution_manager: ItemStorageABC["GraphExecutionState"]):
+        self.graph_execution_manager = graph_execution_manager
+        # {graph_id => NodeLog}
+        self._stats: Dict[str, NodeLog] = {}
+
+    class StatsContext:
+        def __init__(self, invocation: BaseInvocation, graph_id: str, collector: "InvocationStatsServiceBase"):
+            self.invocation = invocation
+            self.collector = collector
+            self.graph_id = graph_id
+            self.start_time = 0
+
+        def __enter__(self):
+            self.start_time = time.time()
+            if torch.cuda.is_available():
+                torch.cuda.reset_peak_memory_stats()
+
+        def __exit__(self, *args):
+            self.collector.update_invocation_stats(
+                self.graph_id,
+                self.invocation.type,
+                time.time() - self.start_time,
+                torch.cuda.max_memory_allocated() / 1e9 if torch.cuda.is_available() else 0.0,
+            )
+
+    def collect_stats(
+        self,
+        invocation: BaseInvocation,
+        graph_execution_state_id: str,
+    ) -> StatsContext:
+        """
+        Return a context object that will capture the statistics.
+        :param invocation: BaseInvocation object from the current graph.
+        :param graph_execution_state: GraphExecutionState object from the current session.
+        """
+        if not self._stats.get(graph_execution_state_id):  # first time we're seeing this
+            self._stats[graph_execution_state_id] = NodeLog()
+        return self.StatsContext(invocation, graph_execution_state_id, self)
+
+    def reset_all_stats(self):
+        """Zero all statistics"""
+        self._stats = {}
+
+    def reset_stats(self, graph_execution_id: str):
+        """Zero the statistics for the indicated graph."""
+        try:
+            self._stats.pop(graph_execution_id)
+        except KeyError:
+            logger.warning(f"Attempted to clear statistics for unknown graph {graph_execution_id}")
+
+    def update_invocation_stats(self, graph_id: str, invocation_type: str, time_used: float, vram_used: float):
+        """
+        Add timing information on execution of a node. Usually
+        used internally.
+        :param graph_id: ID of the graph that is currently executing
+        :param invocation_type: String literal type of the node
+        :param time_used: Floating point seconds used by node's exection
+        """
+        if not self._stats[graph_id].nodes.get(invocation_type):
+            self._stats[graph_id].nodes[invocation_type] = NodeStats()
+        stats = self._stats[graph_id].nodes[invocation_type]
+        stats.calls += 1
+        stats.time_used += time_used
+        stats.max_vram = max(stats.max_vram, vram_used)
+
+    def log_stats(self):
+        """
+        Send the statistics to the system logger at the info level.
+        Stats will only be printed if when the execution of the graph
+        is complete.
+        """
+        completed = set()
+        for graph_id, node_log in self._stats.items():
+            current_graph_state = self.graph_execution_manager.get(graph_id)
+            if not current_graph_state.is_complete():
+                continue
+
+            total_time = 0
+            logger.info(f"Graph stats: {graph_id}")
+            logger.info("Node                 Calls    Seconds VRAM Used")
+            for node_type, stats in self._stats[graph_id].nodes.items():
+                logger.info(f"{node_type:<20} {stats.calls:>5}   {stats.time_used:7.3f}s     {stats.max_vram:4.2f}G")
+                total_time += stats.time_used
+
+            logger.info(f"TOTAL GRAPH EXECUTION TIME:  {total_time:7.3f}s")
+            if torch.cuda.is_available():
+                logger.info("Current VRAM utilization " + "%4.2fG" % (torch.cuda.memory_allocated() / 1e9))
+
+            completed.add(graph_id)
+
+        for graph_id in completed:
+            del self._stats[graph_id]

invokeai/app/services/model_manager_service.py

@@ -3,9 +3,10 @@
 from __future__ import annotations
 
 from abc import ABC, abstractmethod
+from logging import Logger
 from pathlib import Path
 from pydantic import Field
-from typing import Optional, Union, Callable, List, Tuple, TYPE_CHECKING
+from typing import Literal, Optional, Union, Callable, List, Tuple, TYPE_CHECKING
 from types import ModuleType
 
 from invokeai.backend.model_management import (
@@ -193,7 +194,7 @@ class ModelManagerServiceBase(ABC):
         self,
         model_name: str,
         base_model: BaseModelType,
-        model_type: Union[ModelType.Main, ModelType.Vae],
+        model_type: Literal[ModelType.Main, ModelType.Vae],
     ) -> AddModelResult:
         """
         Convert a checkpoint file into a diffusers folder, deleting the cached
@@ -292,7 +293,7 @@ class ModelManagerService(ModelManagerServiceBase):
     def __init__(
         self,
         config: InvokeAIAppConfig,
-        logger: ModuleType,
+        logger: Logger,
     ):
         """
         Initialize with the path to the models.yaml config file.
@@ -396,7 +397,7 @@ class ModelManagerService(ModelManagerServiceBase):
             model_type,
         )
 
-    def model_info(self, model_name: str, base_model: BaseModelType, model_type: ModelType) -> dict:
+    def model_info(self, model_name: str, base_model: BaseModelType, model_type: ModelType) -> Union[dict, None]:
         """
         Given a model name returns a dict-like (OmegaConf) object describing it.
         """
@@ -416,7 +417,7 @@ class ModelManagerService(ModelManagerServiceBase):
         """
         return self.mgr.list_models(base_model, model_type)
 
-    def list_model(self, model_name: str, base_model: BaseModelType, model_type: ModelType) -> dict:
+    def list_model(self, model_name: str, base_model: BaseModelType, model_type: ModelType) -> Union[dict, None]:
         """
         Return information about the model using the same format as list_models()
         """
@@ -429,7 +430,7 @@ class ModelManagerService(ModelManagerServiceBase):
         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.
@@ -478,7 +479,7 @@ class ModelManagerService(ModelManagerServiceBase):
         self,
         model_name: str,
         base_model: BaseModelType,
-        model_type: Union[ModelType.Main, ModelType.Vae],
+        model_type: Literal[ModelType.Main, ModelType.Vae],
         convert_dest_directory: Optional[Path] = Field(
             default=None, description="Optional directory location for merged model"
         ),
@@ -573,9 +574,9 @@ class ModelManagerService(ModelManagerServiceBase):
             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,
+        alpha: float = 0.5,
         interp: Optional[MergeInterpolationMethod] = None,
-        force: Optional[bool] = False,
+        force: bool = False,
         merge_dest_directory: Optional[Path] = Field(
             default=None, description="Optional directory location for merged model"
         ),
@@ -633,8 +634,8 @@ class ModelManagerService(ModelManagerServiceBase):
         model_name: str,
         base_model: BaseModelType,
         model_type: ModelType,
-        new_name: str = None,
-        new_base: BaseModelType = None,
+        new_name: Optional[str] = None,
+        new_base: Optional[BaseModelType] = None,
     ):
         """
         Rename the indicated model. Can provide a new name and/or a new base.

invokeai/app/services/models/board_image.py

@@ -0,0 +1,8 @@
+from pydantic import Field
+
+from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
+
+
+class BoardImage(BaseModelExcludeNull):
+    board_id: str = Field(description="The id of the board")
+    image_name: str = Field(description="The name of the image")

invokeai/app/services/models/board_record.py

@@ -1,10 +1,11 @@
 from typing import Optional, Union
 from datetime import datetime
-from pydantic import BaseModel, Extra, Field, StrictBool, StrictStr
+from pydantic import Field
 from invokeai.app.util.misc import get_iso_timestamp
+from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
 
 
-class BoardRecord(BaseModel):
+class BoardRecord(BaseModelExcludeNull):
     """Deserialized board record."""
 
     board_id: str = Field(description="The unique ID of the board.")

invokeai/app/services/models/image_record.py

@@ -1,13 +1,14 @@
 import datetime
 from typing import Optional, Union
 
-from pydantic import BaseModel, Extra, Field, StrictBool, StrictStr
+from pydantic import Extra, Field, StrictBool, StrictStr
 
 from invokeai.app.models.image import ImageCategory, ResourceOrigin
 from invokeai.app.util.misc import get_iso_timestamp
+from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
 
 
-class ImageRecord(BaseModel):
+class ImageRecord(BaseModelExcludeNull):
     """Deserialized image record without metadata."""
 
     image_name: str = Field(description="The unique name of the image.")
@@ -40,7 +41,7 @@ class ImageRecord(BaseModel):
     """The node ID that generated this image, if it is a generated image."""
 
 
-class ImageRecordChanges(BaseModel, extra=Extra.forbid):
+class ImageRecordChanges(BaseModelExcludeNull, extra=Extra.forbid):
     """A set of changes to apply to an image record.
 
     Only limited changes are valid:
@@ -60,7 +61,7 @@ class ImageRecordChanges(BaseModel, extra=Extra.forbid):
     """The image's new `is_intermediate` flag."""
 
 
-class ImageUrlsDTO(BaseModel):
+class ImageUrlsDTO(BaseModelExcludeNull):
     """The URLs for an image and its thumbnail."""
 
     image_name: str = Field(description="The unique name of the image.")
@@ -76,11 +77,15 @@ class ImageDTO(ImageRecord, ImageUrlsDTO):
 
     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."""
+
     pass
 
 
 def image_record_to_dto(
-    image_record: ImageRecord, image_url: str, thumbnail_url: str, board_id: Optional[str]
+    image_record: ImageRecord,
+    image_url: str,
+    thumbnail_url: str,
+    board_id: Optional[str],
 ) -> ImageDTO:
     """Converts an image record to an image DTO."""
     return ImageDTO(

invokeai/app/services/processor.py

@@ -1,14 +1,15 @@
 import time
 import traceback
-from threading import Event, Thread, BoundedSemaphore
-
-from ..invocations.baseinvocation import InvocationContext
-from .invocation_queue import InvocationQueueItem
-from .invoker import InvocationProcessorABC, Invoker
-from ..models.exceptions import CanceledException
+from threading import BoundedSemaphore, Event, Thread
 
 import invokeai.backend.util.logging as logger
 
+from ..invocations.baseinvocation import InvocationContext
+from ..models.exceptions import CanceledException
+from .invocation_queue import InvocationQueueItem
+from .invocation_stats import InvocationStatsServiceBase
+from .invoker import InvocationProcessorABC, Invoker
+
 
 class DefaultInvocationProcessor(InvocationProcessorABC):
     __invoker_thread: Thread
@@ -35,6 +36,8 @@ class DefaultInvocationProcessor(InvocationProcessorABC):
     def __process(self, stop_event: Event):
         try:
             self.__threadLimit.acquire()
+            statistics: InvocationStatsServiceBase = self.__invoker.services.performance_statistics
+
             while not stop_event.is_set():
                 try:
                     queue_item: InvocationQueueItem = self.__invoker.services.queue.get()
@@ -83,35 +86,38 @@ class DefaultInvocationProcessor(InvocationProcessorABC):
 
                 # Invoke
                 try:
-                    outputs = invocation.invoke(
-                        InvocationContext(
-                            services=self.__invoker.services,
-                            graph_execution_state_id=graph_execution_state.id,
+                    with statistics.collect_stats(invocation, graph_execution_state.id):
+                        outputs = invocation.invoke(
+                            InvocationContext(
+                                services=self.__invoker.services,
+                                graph_execution_state_id=graph_execution_state.id,
+                            )
                         )
-                    )
 
-                    # Check queue to see if this is canceled, and skip if so
-                    if self.__invoker.services.queue.is_canceled(graph_execution_state.id):
-                        continue
+                        # Check queue to see if this is canceled, and skip if so
+                        if self.__invoker.services.queue.is_canceled(graph_execution_state.id):
+                            continue
 
-                    # Save outputs and history
-                    graph_execution_state.complete(invocation.id, outputs)
+                        # Save outputs and history
+                        graph_execution_state.complete(invocation.id, outputs)
 
-                    # Save the state changes
-                    self.__invoker.services.graph_execution_manager.set(graph_execution_state)
+                        # Save the state changes
+                        self.__invoker.services.graph_execution_manager.set(graph_execution_state)
 
-                    # Send complete event
-                    self.__invoker.services.events.emit_invocation_complete(
-                        graph_execution_state_id=graph_execution_state.id,
-                        node=invocation.dict(),
-                        source_node_id=source_node_id,
-                        result=outputs.dict(),
-                    )
+                        # Send complete event
+                        self.__invoker.services.events.emit_invocation_complete(
+                            graph_execution_state_id=graph_execution_state.id,
+                            node=invocation.dict(),
+                            source_node_id=source_node_id,
+                            result=outputs.dict(),
+                        )
+                    statistics.log_stats()
 
                 except KeyboardInterrupt:
                     pass
 
                 except CanceledException:
+                    statistics.reset_stats(graph_execution_state.id)
                     pass
 
                 except Exception as e:
@@ -133,7 +139,7 @@ class DefaultInvocationProcessor(InvocationProcessorABC):
                         error_type=e.__class__.__name__,
                         error=error,
                     )
-
+                    statistics.reset_stats(graph_execution_state.id)
                     pass
 
                 # Check queue to see if this is canceled, and skip if so

invokeai/app/services/urls.py

@@ -20,6 +20,6 @@ class LocalUrlService(UrlServiceBase):
 
         # These paths are determined by the routes in invokeai/app/api/routers/images.py
         if thumbnail:
-            return f"{self._base_url}/images/{image_basename}/thumbnail"
+            return f"{self._base_url}/images/i/{image_basename}/thumbnail"
 
-        return f"{self._base_url}/images/{image_basename}/full"
+        return f"{self._base_url}/images/i/{image_basename}/full"

invokeai/app/util/misc.py

@@ -18,5 +18,5 @@ SEED_MAX = np.iinfo(np.uint32).max
 
 
 def get_random_seed():
-    rng = np.random.default_rng(seed=0)
+    rng = np.random.default_rng(seed=None)
     return int(rng.integers(0, SEED_MAX))

invokeai/app/util/model_exclude_null.py

@@ -0,0 +1,23 @@
+from typing import Any
+from pydantic import BaseModel
+
+
+"""
+We want to exclude null values from objects that make their way to the client.
+
+Unfortunately there is no built-in way to do this in pydantic, so we need to override the default
+dict method to do this.
+
+From https://github.com/tiangolo/fastapi/discussions/8882#discussioncomment-5154541
+"""
+
+
+class BaseModelExcludeNull(BaseModel):
+    def dict(self, *args, **kwargs) -> dict[str, Any]:
+        """
+        Override the default dict method to exclude None values in the response
+        """
+        kwargs.pop("exclude_none", None)
+        return super().dict(*args, exclude_none=True, **kwargs)
+
+    pass

invokeai/backend/install/check_root.py

@@ -12,16 +12,17 @@ def check_invokeai_root(config: InvokeAIAppConfig):
         assert config.model_conf_path.exists(), f"{config.model_conf_path} not found"
         assert config.db_path.parent.exists(), f"{config.db_path.parent} not found"
         assert config.models_path.exists(), f"{config.models_path} not found"
-        for model in [
-            "CLIP-ViT-bigG-14-laion2B-39B-b160k",
-            "bert-base-uncased",
-            "clip-vit-large-patch14",
-            "sd-vae-ft-mse",
-            "stable-diffusion-2-clip",
-            "stable-diffusion-safety-checker",
-        ]:
-            path = config.models_path / f"core/convert/{model}"
-            assert path.exists(), f"{path} is missing"
+        if not config.ignore_missing_core_models:
+            for model in [
+                "CLIP-ViT-bigG-14-laion2B-39B-b160k",
+                "bert-base-uncased",
+                "clip-vit-large-patch14",
+                "sd-vae-ft-mse",
+                "stable-diffusion-2-clip",
+                "stable-diffusion-safety-checker",
+            ]:
+                path = config.models_path / f"core/convert/{model}"
+                assert path.exists(), f"{path} is missing"
     except Exception as e:
         print()
         print(f"An exception has occurred: {str(e)}")
@@ -32,5 +33,10 @@ def check_invokeai_root(config: InvokeAIAppConfig):
         print(
             '** From the command line, activate the virtual environment and run "invokeai-configure --yes --skip-sd-weights" **'
         )
+        print(
+            '** (To skip this check completely, add "--ignore_missing_core_models" to your CLI args. Not installing '
+            "these core models will prevent the loading of some or all .safetensors and .ckpt files. However, you can "
+            "always come back and install these core models in the future.)"
+        )
         input("Press any key to continue...")
         sys.exit(0)

invokeai/backend/install/invokeai_configure.py

@@ -181,7 +181,7 @@ def download_with_progress_bar(model_url: str, model_dest: str, label: str = "th
 
 
 def download_conversion_models():
-    target_dir = config.root_path / "models/core/convert"
+    target_dir = config.models_path / "core/convert"
     kwargs = dict()  # for future use
     try:
         logger.info("Downloading core tokenizers and text encoders")

invokeai/backend/install/model_install_backend.py

@@ -7,11 +7,13 @@ import warnings
 from dataclasses import dataclass, field
 from pathlib import Path
 from tempfile import TemporaryDirectory
-from typing import List, Dict, Callable, Union, Set
+from typing import Optional, List, Dict, Callable, Union, Set
 
 import requests
 from diffusers import DiffusionPipeline
 from diffusers import logging as dlogging
+import onnx
+import torch
 from huggingface_hub import hf_hub_url, HfFolder, HfApi
 from omegaconf import OmegaConf
 from tqdm import tqdm
@@ -22,6 +24,7 @@ from invokeai.app.services.config import InvokeAIAppConfig
 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 invokeai.backend.util.devices import torch_dtype, choose_torch_device
 from ..util.logging import InvokeAILogger
 
 warnings.filterwarnings("ignore")
@@ -86,8 +89,8 @@ class ModelLoadInfo:
     name: str
     model_type: ModelType
     base_type: BaseModelType
-    path: Path = None
-    repo_id: str = None
+    path: Optional[Path] = None
+    repo_id: Optional[str] = None
     description: str = ""
     installed: bool = False
     recommended: bool = False
@@ -98,9 +101,9 @@ class ModelInstall(object):
     def __init__(
         self,
         config: InvokeAIAppConfig,
-        prediction_type_helper: Callable[[Path], SchedulerPredictionType] = None,
-        model_manager: ModelManager = None,
-        access_token: str = None,
+        prediction_type_helper: Optional[Callable[[Path], SchedulerPredictionType]] = None,
+        model_manager: Optional[ModelManager] = None,
+        access_token: Optional[str] = None,
     ):
         self.config = config
         self.mgr = model_manager or ModelManager(config.model_conf_path)
@@ -128,7 +131,9 @@ class ModelInstall(object):
             model_dict[key] = ModelLoadInfo(**value)
 
         # supplement with entries in models.yaml
-        installed_models = self.mgr.list_models()
+        installed_models = [x for x in self.mgr.list_models()]
+        # suppresses autoloaded models
+        # installed_models = [x for x in self.mgr.list_models() if not self._is_autoloaded(x)]
 
         for md in installed_models:
             base = md["base_model"]
@@ -147,6 +152,17 @@ class ModelInstall(object):
                 )
         return {x: model_dict[x] for x in sorted(model_dict.keys(), key=lambda y: model_dict[y].name.lower())}
 
+    def _is_autoloaded(self, model_info: dict) -> bool:
+        path = model_info.get("path")
+        if not path:
+            return False
+        for autodir in ["autoimport_dir", "lora_dir", "embedding_dir", "controlnet_dir"]:
+            if autodir_path := getattr(self.config, autodir):
+                autodir_path = self.config.root_path / autodir_path
+                if Path(path).is_relative_to(autodir_path):
+                    return True
+        return False
+
     def list_models(self, model_type):
         installed = self.mgr.list_models(model_type=model_type)
         print(f"Installed models of type `{model_type}`:")
@@ -273,6 +289,7 @@ class ModelInstall(object):
                 logger.error(f"Unable to download {url}. Skipping.")
             info = ModelProbe().heuristic_probe(location)
             dest = self.config.models_path / info.base_type.value / info.model_type.value / location.name
+            dest.parent.mkdir(parents=True, exist_ok=True)
             models_path = shutil.move(location, dest)
 
         # staged version will be garbage-collected at this time
@@ -290,6 +307,8 @@ class ModelInstall(object):
             staging = Path(staging)
             if "model_index.json" in files:
                 location = self._download_hf_pipeline(repo_id, staging)  # pipeline
+            elif "unet/model.onnx" in files:
+                location = self._download_hf_model(repo_id, files, staging)
             else:
                 for suffix in ["safetensors", "bin"]:
                     if f"pytorch_lora_weights.{suffix}" in files:
@@ -346,7 +365,7 @@ class ModelInstall(object):
             if key in self.datasets:
                 description = self.datasets[key].get("description") or description
 
-        rel_path = self.relative_to_root(path)
+        rel_path = self.relative_to_root(path, self.config.models_path)
 
         attributes = dict(
             path=str(rel_path),
@@ -354,7 +373,7 @@ class ModelInstall(object):
             model_format=info.format,
         )
         legacy_conf = None
-        if info.model_type == ModelType.Main:
+        if info.model_type == ModelType.Main or info.model_type == ModelType.ONNX:
             attributes.update(
                 dict(
                     variant=info.variant_type,
@@ -386,8 +405,8 @@ class ModelInstall(object):
             attributes.update(dict(config=str(legacy_conf)))
         return attributes
 
-    def relative_to_root(self, path: Path) -> Path:
-        root = self.config.root_path
+    def relative_to_root(self, path: Path, root: Optional[Path] = None) -> Path:
+        root = root or self.config.root_path
         if path.is_relative_to(root):
             return path.relative_to(root)
         else:
@@ -399,15 +418,25 @@ class ModelInstall(object):
         does a save_pretrained() to the indicated staging area.
         """
         _, name = repo_id.split("/")
-        revisions = ["fp16", "main"] if self.config.precision == "float16" else ["main"]
+        precision = torch_dtype(choose_torch_device())
+        variants = ["fp16", None] if precision == torch.float16 else [None, "fp16"]
+
         model = None
-        for revision in revisions:
+        for variant in variants:
             try:
-                model = DiffusionPipeline.from_pretrained(repo_id, revision=revision, safety_checker=None)
-            except:  # most errors are due to fp16 not being present. Fix this to catch other errors
-                pass
+                model = DiffusionPipeline.from_pretrained(
+                    repo_id,
+                    variant=variant,
+                    torch_dtype=precision,
+                    safety_checker=None,
+                )
+            except Exception as e:  # most errors are due to fp16 not being present. Fix this to catch other errors
+                if "fp16" not in str(e):
+                    print(e)
+
             if model:
                 break
+
         if not model:
             logger.error(f"Diffusers model {repo_id} could not be downloaded. Skipping.")
             return None
@@ -419,8 +448,13 @@ class ModelInstall(object):
         location = staging / name
         paths = list()
         for filename in files:
+            filePath = Path(filename)
             p = hf_download_with_resume(
-                repo_id, model_dir=location, model_name=filename, access_token=self.access_token
+                repo_id,
+                model_dir=location / filePath.parent,
+                model_name=filePath.name,
+                access_token=self.access_token,
+                subfolder=filePath.parent,
             )
             if p:
                 paths.append(p)
@@ -468,11 +502,12 @@ def hf_download_with_resume(
     model_name: str,
     model_dest: Path = None,
     access_token: str = None,
+    subfolder: str = None,
 ) -> Path:
     model_dest = model_dest or Path(os.path.join(model_dir, model_name))
     os.makedirs(model_dir, exist_ok=True)
 
-    url = hf_hub_url(repo_id, model_name)
+    url = hf_hub_url(repo_id, model_name, subfolder=subfolder)
 
     header = {"Authorization": f"Bearer {access_token}"} if access_token else {}
     open_mode = "wb"

invokeai/backend/model_management/__init__.py

@@ -3,6 +3,7 @@ Initialization file for invokeai.backend.model_management
 """
 from .model_manager import ModelManager, ModelInfo, AddModelResult, SchedulerPredictionType
 from .model_cache import ModelCache
+from .lora import ModelPatcher, ONNXModelPatcher
 from .models import (
     BaseModelType,
     ModelType,
@@ -12,3 +13,4 @@ from .models import (
     DuplicateModelException,
 )
 from .model_merge import ModelMerger, MergeInterpolationMethod
+from .lora import ModelPatcher

invokeai/backend/model_management/convert_ckpt_to_diffusers.py

@@ -63,7 +63,7 @@ from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionS
 from diffusers.pipelines.stable_diffusion.stable_unclip_image_normalizer import StableUnCLIPImageNormalizer
 
 from invokeai.backend.util.logging import InvokeAILogger
-from invokeai.app.services.config import InvokeAIAppConfig, MODEL_CORE
+from invokeai.app.services.config import InvokeAIAppConfig
 
 from picklescan.scanner import scan_file_path
 from .models import BaseModelType, ModelVariantType
@@ -81,7 +81,7 @@ if is_accelerate_available():
     from accelerate.utils import set_module_tensor_to_device
 
 logger = InvokeAILogger.getLogger(__name__)
-CONVERT_MODEL_ROOT = InvokeAIAppConfig.get_config().root_path / MODEL_CORE / "convert"
+CONVERT_MODEL_ROOT = InvokeAIAppConfig.get_config().models_path / "core/convert"
 
 
 def shave_segments(path, n_shave_prefix_segments=1):
@@ -1070,7 +1070,7 @@ def convert_controlnet_checkpoint(
     extract_ema,
     use_linear_projection=None,
     cross_attention_dim=None,
-    precision: torch.dtype = torch.float32,
+    precision: Optional[torch.dtype] = None,
 ):
     ctrlnet_config = create_unet_diffusers_config(original_config, image_size=image_size, controlnet=True)
     ctrlnet_config["upcast_attention"] = upcast_attention
@@ -1111,7 +1111,6 @@ def convert_controlnet_checkpoint(
     return controlnet.to(precision)
 
 
-# TO DO - PASS PRECISION
 def download_from_original_stable_diffusion_ckpt(
     checkpoint_path: str,
     model_version: BaseModelType,
@@ -1121,7 +1120,7 @@ def download_from_original_stable_diffusion_ckpt(
     prediction_type: str = None,
     model_type: str = None,
     extract_ema: bool = False,
-    precision: torch.dtype = torch.float32,
+    precision: Optional[torch.dtype] = None,
     scheduler_type: str = "pndm",
     num_in_channels: Optional[int] = None,
     upcast_attention: Optional[bool] = None,
@@ -1194,6 +1193,8 @@ def download_from_original_stable_diffusion_ckpt(
             [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer)
             to use. If this parameter is `None`, the function will load a new instance of [CLIPTokenizer] by itself, if
             needed.
+        precision (`torch.dtype`, *optional*, defauts to `None`):
+            If not provided the precision will be set to the precision of the original file.
         return: A StableDiffusionPipeline object representing the passed-in `.ckpt`/`.safetensors` file.
     """
 
@@ -1252,6 +1253,10 @@ def download_from_original_stable_diffusion_ckpt(
 
     logger.debug(f"model_type = {model_type}; original_config_file = {original_config_file}")
 
+    precision_probing_key = "model.diffusion_model.input_blocks.0.0.bias"
+    logger.debug(f"original checkpoint precision == {checkpoint[precision_probing_key].dtype}")
+    precision = precision or checkpoint[precision_probing_key].dtype
+
     if original_config_file is None:
         key_name_v2_1 = "model.diffusion_model.input_blocks.2.1.transformer_blocks.0.attn2.to_k.weight"
         key_name_sd_xl_base = "conditioner.embedders.1.model.transformer.resblocks.9.mlp.c_proj.bias"
@@ -1279,9 +1284,12 @@ def download_from_original_stable_diffusion_ckpt(
         original_config_file = BytesIO(requests.get(config_url).content)
 
     original_config = OmegaConf.load(original_config_file)
+    if original_config["model"]["params"].get("use_ema") is not None:
+        extract_ema = original_config["model"]["params"]["use_ema"]
+
     if (
         model_version == BaseModelType.StableDiffusion2
-        and original_config["model"]["params"]["parameterization"] == "v"
+        and original_config["model"]["params"].get("parameterization") == "v"
     ):
         prediction_type = "v_prediction"
         upcast_attention = True
@@ -1447,7 +1455,7 @@ def download_from_original_stable_diffusion_ckpt(
             if controlnet:
                 pipe = pipeline_class(
                     vae=vae.to(precision),
-                    text_encoder=text_model,
+                    text_encoder=text_model.to(precision),
                     tokenizer=tokenizer,
                     unet=unet.to(precision),
                     scheduler=scheduler,
@@ -1459,7 +1467,7 @@ def download_from_original_stable_diffusion_ckpt(
             else:
                 pipe = pipeline_class(
                     vae=vae.to(precision),
-                    text_encoder=text_model,
+                    text_encoder=text_model.to(precision),
                     tokenizer=tokenizer,
                     unet=unet.to(precision),
                     scheduler=scheduler,
@@ -1484,8 +1492,8 @@ def download_from_original_stable_diffusion_ckpt(
                     image_noising_scheduler=image_noising_scheduler,
                     # regular denoising components
                     tokenizer=tokenizer,
-                    text_encoder=text_model,
-                    unet=unet,
+                    text_encoder=text_model.to(precision),
+                    unet=unet.to(precision),
                     scheduler=scheduler,
                     # vae
                     vae=vae,
@@ -1560,7 +1568,7 @@ def download_from_original_stable_diffusion_ckpt(
         if controlnet:
             pipe = pipeline_class(
                 vae=vae.to(precision),
-                text_encoder=text_model,
+                text_encoder=text_model.to(precision),
                 tokenizer=tokenizer,
                 unet=unet.to(precision),
                 controlnet=controlnet,
@@ -1571,7 +1579,7 @@ def download_from_original_stable_diffusion_ckpt(
         else:
             pipe = pipeline_class(
                 vae=vae.to(precision),
-                text_encoder=text_model,
+                text_encoder=text_model.to(precision),
                 tokenizer=tokenizer,
                 unet=unet.to(precision),
                 scheduler=scheduler,
@@ -1594,9 +1602,9 @@ def download_from_original_stable_diffusion_ckpt(
 
             pipe = StableDiffusionXLPipeline(
                 vae=vae.to(precision),
-                text_encoder=text_encoder,
+                text_encoder=text_encoder.to(precision),
                 tokenizer=tokenizer,
-                text_encoder_2=text_encoder_2,
+                text_encoder_2=text_encoder_2.to(precision),
                 tokenizer_2=tokenizer_2,
                 unet=unet.to(precision),
                 scheduler=scheduler,
@@ -1639,7 +1647,7 @@ def download_controlnet_from_original_ckpt(
     original_config_file: str,
     image_size: int = 512,
     extract_ema: bool = False,
-    precision: torch.dtype = torch.float32,
+    precision: Optional[torch.dtype] = None,
     num_in_channels: Optional[int] = None,
     upcast_attention: Optional[bool] = None,
     device: str = None,
@@ -1680,6 +1688,12 @@ def download_controlnet_from_original_ckpt(
     while "state_dict" in checkpoint:
         checkpoint = checkpoint["state_dict"]
 
+    # use original precision
+    precision_probing_key = "input_blocks.0.0.bias"
+    ckpt_precision = checkpoint[precision_probing_key].dtype
+    logger.debug(f"original controlnet precision = {ckpt_precision}")
+    precision = precision or ckpt_precision
+
     original_config = OmegaConf.load(original_config_file)
 
     if num_in_channels is not None:
@@ -1699,7 +1713,7 @@ def download_controlnet_from_original_ckpt(
         cross_attention_dim=cross_attention_dim,
     )
 
-    return controlnet
+    return controlnet.to(precision)
 
 
 def convert_ldm_vae_to_diffusers(checkpoint, vae_config: DictConfig, image_size: int) -> AutoencoderKL:

invokeai/backend/model_management/lora.py

@@ -6,427 +6,20 @@ 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 onnx import numpy_helper
+from onnxruntime import OrtValue
+import numpy as np
+
 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]
-    # alpha: Optional[float]
-    # bias: Optional[torch.Tensor]
-    # layer_key: str
-
-    # @property
-    # def scale(self):
-    #    return self.alpha / self.rank if (self.alpha and self.rank) else 1.0
-
-    def __init__(
-        self,
-        layer_key: str,
-        values: dict,
-    ):
-        if "alpha" in values:
-            self.alpha = values["alpha"].item()
-        else:
-            self.alpha = None
-
-        if "bias_indices" in values and "bias_values" in values and "bias_size" in values:
-            self.bias = torch.sparse_coo_tensor(
-                values["bias_indices"],
-                values["bias_values"],
-                tuple(values["bias_size"]),
-            )
-
-        else:
-            self.bias = None
-
-        self.rank = None  # set in layer implementation
-        self.layer_key = layer_key
-
-    def forward(
-        self,
-        module: torch.nn.Module,
-        input_h: Any,  # for real looks like Tuple[torch.nn.Tensor] but not sure
-        multiplier: float,
-    ):
-        if type(module) == torch.nn.Conv2d:
-            op = torch.nn.functional.conv2d
-            extra_args = dict(
-                stride=module.stride,
-                padding=module.padding,
-                dilation=module.dilation,
-                groups=module.groups,
-            )
-
-        else:
-            op = torch.nn.functional.linear
-            extra_args = {}
-
-        weight = self.get_weight()
-
-        bias = self.bias if self.bias is not None else 0
-        scale = self.alpha / self.rank if (self.alpha and self.rank) else 1.0
-        return (
-            op(
-                *input_h,
-                (weight + bias).view(module.weight.shape),
-                None,
-                **extra_args,
-            )
-            * multiplier
-            * scale
-        )
-
-    def get_weight(self):
-        raise NotImplementedError()
-
-    def calc_size(self) -> int:
-        model_size = 0
-        for val in [self.bias]:
-            if val is not None:
-                model_size += val.nelement() * val.element_size()
-        return model_size
-
-    def to(
-        self,
-        device: Optional[torch.device] = None,
-        dtype: Optional[torch.dtype] = None,
-    ):
-        if self.bias is not None:
-            self.bias = self.bias.to(device=device, dtype=dtype)
-
-
-# TODO: find and debug lora/locon with bias
-class LoRALayer(LoRALayerBase):
-    # up: torch.Tensor
-    # mid: Optional[torch.Tensor]
-    # down: torch.Tensor
-
-    def __init__(
-        self,
-        layer_key: str,
-        values: dict,
-    ):
-        super().__init__(layer_key, values)
-
-        self.up = values["lora_up.weight"]
-        self.down = values["lora_down.weight"]
-        if "lora_mid.weight" in values:
-            self.mid = values["lora_mid.weight"]
-        else:
-            self.mid = None
-
-        self.rank = self.down.shape[0]
-
-    def get_weight(self):
-        if self.mid is not None:
-            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)
-
-        return weight
-
-    def calc_size(self) -> int:
-        model_size = super().calc_size()
-        for val in [self.up, self.mid, self.down]:
-            if val is not None:
-                model_size += val.nelement() * val.element_size()
-        return model_size
-
-    def to(
-        self,
-        device: Optional[torch.device] = None,
-        dtype: Optional[torch.dtype] = None,
-    ):
-        super().to(device=device, dtype=dtype)
-
-        self.up = self.up.to(device=device, dtype=dtype)
-        self.down = self.down.to(device=device, dtype=dtype)
-
-        if self.mid is not None:
-            self.mid = self.mid.to(device=device, dtype=dtype)
-
-
-class LoHALayer(LoRALayerBase):
-    # w1_a: torch.Tensor
-    # w1_b: torch.Tensor
-    # w2_a: torch.Tensor
-    # w2_b: torch.Tensor
-    # t1: Optional[torch.Tensor] = None
-    # t2: Optional[torch.Tensor] = None
-
-    def __init__(
-        self,
-        layer_key: str,
-        values: dict,
-    ):
-        super().__init__(layer_key, values)
-
-        self.w1_a = values["hada_w1_a"]
-        self.w1_b = values["hada_w1_b"]
-        self.w2_a = values["hada_w2_a"]
-        self.w2_b = values["hada_w2_b"]
-
-        if "hada_t1" in values:
-            self.t1 = values["hada_t1"]
-        else:
-            self.t1 = None
-
-        if "hada_t2" in values:
-            self.t2 = values["hada_t2"]
-        else:
-            self.t2 = None
-
-        self.rank = self.w1_b.shape[0]
-
-    def get_weight(self):
-        if self.t1 is None:
-            weight = (self.w1_a @ self.w1_b) * (self.w2_a @ self.w2_b)
-
-        else:
-            rebuild1 = torch.einsum("i j k l, j r, i p -> p r k l", self.t1, self.w1_b, self.w1_a)
-            rebuild2 = torch.einsum("i j k l, j r, i p -> p r k l", self.t2, self.w2_b, self.w2_a)
-            weight = rebuild1 * rebuild2
-
-        return weight
-
-    def calc_size(self) -> int:
-        model_size = super().calc_size()
-        for val in [self.w1_a, self.w1_b, self.w2_a, self.w2_b, self.t1, self.t2]:
-            if val is not None:
-                model_size += val.nelement() * val.element_size()
-        return model_size
-
-    def to(
-        self,
-        device: Optional[torch.device] = None,
-        dtype: Optional[torch.dtype] = None,
-    ):
-        super().to(device=device, dtype=dtype)
-
-        self.w1_a = self.w1_a.to(device=device, dtype=dtype)
-        self.w1_b = self.w1_b.to(device=device, dtype=dtype)
-        if self.t1 is not None:
-            self.t1 = self.t1.to(device=device, dtype=dtype)
-
-        self.w2_a = self.w2_a.to(device=device, dtype=dtype)
-        self.w2_b = self.w2_b.to(device=device, dtype=dtype)
-        if self.t2 is not None:
-            self.t2 = self.t2.to(device=device, dtype=dtype)
-
-
-class LoKRLayer(LoRALayerBase):
-    # w1: Optional[torch.Tensor] = None
-    # w1_a: Optional[torch.Tensor] = None
-    # w1_b: Optional[torch.Tensor] = None
-    # w2: Optional[torch.Tensor] = None
-    # w2_a: Optional[torch.Tensor] = None
-    # w2_b: Optional[torch.Tensor] = None
-    # t2: Optional[torch.Tensor] = None
-
-    def __init__(
-        self,
-        layer_key: str,
-        values: dict,
-    ):
-        super().__init__(layer_key, values)
-
-        if "lokr_w1" in values:
-            self.w1 = values["lokr_w1"]
-            self.w1_a = None
-            self.w1_b = None
-        else:
-            self.w1 = None
-            self.w1_a = values["lokr_w1_a"]
-            self.w1_b = values["lokr_w1_b"]
-
-        if "lokr_w2" in values:
-            self.w2 = values["lokr_w2"]
-            self.w2_a = None
-            self.w2_b = None
-        else:
-            self.w2 = None
-            self.w2_a = values["lokr_w2_a"]
-            self.w2_b = values["lokr_w2_b"]
-
-        if "lokr_t2" in values:
-            self.t2 = values["lokr_t2"]
-        else:
-            self.t2 = None
-
-        if "lokr_w1_b" in values:
-            self.rank = values["lokr_w1_b"].shape[0]
-        elif "lokr_w2_b" in values:
-            self.rank = values["lokr_w2_b"].shape[0]
-        else:
-            self.rank = None  # unscaled
-
-    def get_weight(self):
-        w1 = self.w1
-        if w1 is None:
-            w1 = self.w1_a @ self.w1_b
-
-        w2 = self.w2
-        if w2 is None:
-            if self.t2 is None:
-                w2 = self.w2_a @ self.w2_b
-            else:
-                w2 = torch.einsum("i j k l, i p, j r -> p r k l", self.t2, self.w2_a, self.w2_b)
-
-        if len(w2.shape) == 4:
-            w1 = w1.unsqueeze(2).unsqueeze(2)
-        w2 = w2.contiguous()
-        weight = torch.kron(w1, w2)
-
-        return weight
-
-    def calc_size(self) -> int:
-        model_size = super().calc_size()
-        for val in [self.w1, self.w1_a, self.w1_b, self.w2, self.w2_a, self.w2_b, self.t2]:
-            if val is not None:
-                model_size += val.nelement() * val.element_size()
-        return model_size
-
-    def to(
-        self,
-        device: Optional[torch.device] = None,
-        dtype: Optional[torch.dtype] = None,
-    ):
-        super().to(device=device, dtype=dtype)
-
-        if self.w1 is not None:
-            self.w1 = self.w1.to(device=device, dtype=dtype)
-        else:
-            self.w1_a = self.w1_a.to(device=device, dtype=dtype)
-            self.w1_b = self.w1_b.to(device=device, dtype=dtype)
-
-        if self.w2 is not None:
-            self.w2 = self.w2.to(device=device, dtype=dtype)
-        else:
-            self.w2_a = self.w2_a.to(device=device, dtype=dtype)
-            self.w2_b = self.w2_b.to(device=device, dtype=dtype)
-
-        if self.t2 is not None:
-            self.t2 = self.t2.to(device=device, dtype=dtype)
-
-
-class LoRAModel:  # (torch.nn.Module):
-    _name: str
-    layers: Dict[str, LoRALayer]
-    _device: torch.device
-    _dtype: torch.dtype
-
-    def __init__(
-        self,
-        name: str,
-        layers: Dict[str, LoRALayer],
-        device: torch.device,
-        dtype: torch.dtype,
-    ):
-        self._name = name
-        self._device = device or torch.cpu
-        self._dtype = dtype or torch.float32
-        self.layers = layers
-
-    @property
-    def name(self):
-        return self._name
-
-    @property
-    def device(self):
-        return self._device
-
-    @property
-    def dtype(self):
-        return self._dtype
-
-    def to(
-        self,
-        device: Optional[torch.device] = None,
-        dtype: Optional[torch.dtype] = None,
-    ) -> LoRAModel:
-        # TODO: try revert if exception?
-        for key, layer in self.layers.items():
-            layer.to(device=device, dtype=dtype)
-        self._device = device
-        self._dtype = dtype
-
-    def calc_size(self) -> int:
-        model_size = 0
-        for _, layer in self.layers.items():
-            model_size += layer.calc_size()
-        return model_size
-
-    @classmethod
-    def from_checkpoint(
-        cls,
-        file_path: Union[str, Path],
-        device: Optional[torch.device] = None,
-        dtype: Optional[torch.dtype] = None,
-    ):
-        device = device or torch.device("cpu")
-        dtype = dtype or torch.float32
-
-        if isinstance(file_path, str):
-            file_path = Path(file_path)
-
-        model = cls(
-            device=device,
-            dtype=dtype,
-            name=file_path.stem,  # TODO:
-            layers=dict(),
-        )
-
-        if file_path.suffix == ".safetensors":
-            state_dict = load_file(file_path.absolute().as_posix(), device="cpu")
-        else:
-            state_dict = torch.load(file_path, map_location="cpu")
-
-        state_dict = cls._group_state(state_dict)
-
-        for layer_key, values in state_dict.items():
-            # lora and locon
-            if "lora_down.weight" in values:
-                layer = LoRALayer(layer_key, values)
-
-            # loha
-            elif "hada_w1_b" in values:
-                layer = LoHALayer(layer_key, values)
-
-            # lokr
-            elif "lokr_w1_b" in values or "lokr_w1" in values:
-                layer = LoKRLayer(layer_key, values)
-
-            else:
-                # TODO: diff/ia3/... format
-                print(f">> Encountered unknown lora layer module in {model.name}: {layer_key}")
-                return
-
-            # lower memory consumption by removing already parsed layer values
-            state_dict[layer_key].clear()
-
-            layer.to(device=device, dtype=dtype)
-            model.layers[layer_key] = layer
-
-        return model
-
-    @staticmethod
-    def _group_state(state_dict: dict):
-        state_dict_groupped = dict()
-
-        for key, value in state_dict.items():
-            stem, leaf = key.split(".", 1)
-            if stem not in state_dict_groupped:
-                state_dict_groupped[stem] = dict()
-            state_dict_groupped[stem][leaf] = value
-
-        return state_dict_groupped
-
-
 """
 loras = [
     (lora_model1, 0.7),
@@ -505,6 +98,26 @@ class ModelPatcher:
         with cls.apply_lora(text_encoder, loras, "lora_te_"):
             yield
 
+    @classmethod
+    @contextmanager
+    def apply_sdxl_lora_text_encoder(
+        cls,
+        text_encoder: CLIPTextModel,
+        loras: List[Tuple[LoRAModel, float]],
+    ):
+        with cls.apply_lora(text_encoder, loras, "lora_te1_"):
+            yield
+
+    @classmethod
+    @contextmanager
+    def apply_sdxl_lora_text_encoder2(
+        cls,
+        text_encoder: CLIPTextModel,
+        loras: List[Tuple[LoRAModel, float]],
+    ):
+        with cls.apply_lora(text_encoder, loras, "lora_te2_"):
+            yield
+
     @classmethod
     @contextmanager
     def apply_lora(
@@ -551,7 +164,7 @@ class ModelPatcher:
         cls,
         tokenizer: CLIPTokenizer,
         text_encoder: CLIPTextModel,
-        ti_list: List[Any],
+        ti_list: List[Tuple[str, Any]],
     ) -> Tuple[CLIPTokenizer, TextualInversionManager]:
         init_tokens_count = None
         new_tokens_added = None
@@ -561,27 +174,27 @@ class ModelPatcher:
             ti_manager = TextualInversionManager(ti_tokenizer)
             init_tokens_count = text_encoder.resize_token_embeddings(None).num_embeddings
 
-            def _get_trigger(ti, index):
-                trigger = ti.name
+            def _get_trigger(ti_name, index):
+                trigger = ti_name
                 if index > 0:
                     trigger += f"-!pad-{i}"
                 return f"<{trigger}>"
 
             # modify tokenizer
             new_tokens_added = 0
-            for ti in ti_list:
+            for ti_name, ti in ti_list:
                 for i in range(ti.embedding.shape[0]):
-                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti, i))
+                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti_name, i))
 
             # modify text_encoder
             text_encoder.resize_token_embeddings(init_tokens_count + new_tokens_added)
             model_embeddings = text_encoder.get_input_embeddings()
 
-            for ti in ti_list:
+            for ti_name, ti in ti_list:
                 ti_tokens = []
                 for i in range(ti.embedding.shape[0]):
                     embedding = ti.embedding[i]
-                    trigger = _get_trigger(ti, i)
+                    trigger = _get_trigger(ti_name, i)
 
                     token_id = ti_tokenizer.convert_tokens_to_ids(trigger)
                     if token_id == ti_tokenizer.unk_token_id:
@@ -626,7 +239,6 @@ class ModelPatcher:
 
 
 class TextualInversionModel:
-    name: str
     embedding: torch.Tensor  # [n, 768]|[n, 1280]
 
     @classmethod
@@ -640,7 +252,6 @@ class TextualInversionModel:
             file_path = Path(file_path)
 
         result = cls()  # TODO:
-        result.name = file_path.stem  # TODO:
 
         if file_path.suffix == ".safetensors":
             state_dict = load_file(file_path.absolute().as_posix(), device="cpu")
@@ -698,3 +309,186 @@ class TextualInversionManager(BaseTextualInversionManager):
                 new_token_ids.extend(self.pad_tokens[token_id])
 
         return new_token_ids
+
+
+class ONNXModelPatcher:
+    from .models.base import IAIOnnxRuntimeModel, OnnxRuntimeModel
+
+    @classmethod
+    @contextmanager
+    def apply_lora_unet(
+        cls,
+        unet: OnnxRuntimeModel,
+        loras: List[Tuple[LoRAModel, float]],
+    ):
+        with cls.apply_lora(unet, loras, "lora_unet_"):
+            yield
+
+    @classmethod
+    @contextmanager
+    def apply_lora_text_encoder(
+        cls,
+        text_encoder: OnnxRuntimeModel,
+        loras: List[Tuple[LoRAModel, float]],
+    ):
+        with cls.apply_lora(text_encoder, loras, "lora_te_"):
+            yield
+
+    # based on
+    # https://github.com/ssube/onnx-web/blob/ca2e436f0623e18b4cfe8a0363fcfcf10508acf7/api/onnx_web/convert/diffusion/lora.py#L323
+    @classmethod
+    @contextmanager
+    def apply_lora(
+        cls,
+        model: IAIOnnxRuntimeModel,
+        loras: List[Tuple[LoraModel, float]],
+        prefix: str,
+    ):
+        from .models.base import IAIOnnxRuntimeModel
+
+        if not isinstance(model, IAIOnnxRuntimeModel):
+            raise Exception("Only IAIOnnxRuntimeModel models supported")
+
+        orig_weights = dict()
+
+        try:
+            blended_loras = dict()
+
+            for lora, lora_weight in loras:
+                for layer_key, layer in lora.layers.items():
+                    if not layer_key.startswith(prefix):
+                        continue
+
+                    layer.to(dtype=torch.float32)
+                    layer_key = layer_key.replace(prefix, "")
+                    layer_weight = layer.get_weight().detach().cpu().numpy() * lora_weight
+                    if layer_key is blended_loras:
+                        blended_loras[layer_key] += layer_weight
+                    else:
+                        blended_loras[layer_key] = layer_weight
+
+            node_names = dict()
+            for node in model.nodes.values():
+                node_names[node.name.replace("/", "_").replace(".", "_").lstrip("_")] = node.name
+
+            for layer_key, lora_weight in blended_loras.items():
+                conv_key = layer_key + "_Conv"
+                gemm_key = layer_key + "_Gemm"
+                matmul_key = layer_key + "_MatMul"
+
+                if conv_key in node_names or gemm_key in node_names:
+                    if conv_key in node_names:
+                        conv_node = model.nodes[node_names[conv_key]]
+                    else:
+                        conv_node = model.nodes[node_names[gemm_key]]
+
+                    weight_name = [n for n in conv_node.input if ".weight" in n][0]
+                    orig_weight = model.tensors[weight_name]
+
+                    if orig_weight.shape[-2:] == (1, 1):
+                        if lora_weight.shape[-2:] == (1, 1):
+                            new_weight = orig_weight.squeeze((3, 2)) + lora_weight.squeeze((3, 2))
+                        else:
+                            new_weight = orig_weight.squeeze((3, 2)) + lora_weight
+
+                        new_weight = np.expand_dims(new_weight, (2, 3))
+                    else:
+                        if orig_weight.shape != lora_weight.shape:
+                            new_weight = orig_weight + lora_weight.reshape(orig_weight.shape)
+                        else:
+                            new_weight = orig_weight + lora_weight
+
+                    orig_weights[weight_name] = orig_weight
+                    model.tensors[weight_name] = new_weight.astype(orig_weight.dtype)
+
+                elif matmul_key in node_names:
+                    weight_node = model.nodes[node_names[matmul_key]]
+                    matmul_name = [n for n in weight_node.input if "MatMul" in n][0]
+
+                    orig_weight = model.tensors[matmul_name]
+                    new_weight = orig_weight + lora_weight.transpose()
+
+                    orig_weights[matmul_name] = orig_weight
+                    model.tensors[matmul_name] = new_weight.astype(orig_weight.dtype)
+
+                else:
+                    # warn? err?
+                    pass
+
+            yield
+
+        finally:
+            # restore original weights
+            for name, orig_weight in orig_weights.items():
+                model.tensors[name] = orig_weight
+
+    @classmethod
+    @contextmanager
+    def apply_ti(
+        cls,
+        tokenizer: CLIPTokenizer,
+        text_encoder: IAIOnnxRuntimeModel,
+        ti_list: List[Tuple[str, Any]],
+    ) -> Tuple[CLIPTokenizer, TextualInversionManager]:
+        from .models.base import IAIOnnxRuntimeModel
+
+        if not isinstance(text_encoder, IAIOnnxRuntimeModel):
+            raise Exception("Only IAIOnnxRuntimeModel models supported")
+
+        orig_embeddings = None
+
+        try:
+            ti_tokenizer = copy.deepcopy(tokenizer)
+            ti_manager = TextualInversionManager(ti_tokenizer)
+
+            def _get_trigger(ti_name, index):
+                trigger = ti_name
+                if index > 0:
+                    trigger += f"-!pad-{i}"
+                return f"<{trigger}>"
+
+            # modify tokenizer
+            new_tokens_added = 0
+            for ti_name, ti in ti_list:
+                for i in range(ti.embedding.shape[0]):
+                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti_name, i))
+
+            # modify text_encoder
+            orig_embeddings = text_encoder.tensors["text_model.embeddings.token_embedding.weight"]
+
+            embeddings = np.concatenate(
+                (np.copy(orig_embeddings), np.zeros((new_tokens_added, orig_embeddings.shape[1]))),
+                axis=0,
+            )
+
+            for ti_name, ti in ti_list:
+                ti_tokens = []
+                for i in range(ti.embedding.shape[0]):
+                    embedding = ti.embedding[i].detach().numpy()
+                    trigger = _get_trigger(ti_name, i)
+
+                    token_id = ti_tokenizer.convert_tokens_to_ids(trigger)
+                    if token_id == ti_tokenizer.unk_token_id:
+                        raise RuntimeError(f"Unable to find token id for token '{trigger}'")
+
+                    if embeddings[token_id].shape != embedding.shape:
+                        raise ValueError(
+                            f"Cannot load embedding for {trigger}. It was trained on a model with token dimension {embedding.shape[0]}, but the current model has token dimension {embeddings[token_id].shape[0]}."
+                        )
+
+                    embeddings[token_id] = embedding
+                    ti_tokens.append(token_id)
+
+                if len(ti_tokens) > 1:
+                    ti_manager.pad_tokens[ti_tokens[0]] = ti_tokens[1:]
+
+            text_encoder.tensors["text_model.embeddings.token_embedding.weight"] = embeddings.astype(
+                orig_embeddings.dtype
+            )
+
+            yield ti_tokenizer, ti_manager
+
+        finally:
+            # restore
+            if orig_embeddings is not None:
+                text_encoder.tensors["text_model.embeddings.token_embedding.weight"] = orig_embeddings

invokeai/backend/model_management/model_cache.py

@@ -28,8 +28,6 @@ import torch
 
 import logging
 import invokeai.backend.util.logging as logger
-from invokeai.app.services.config import get_invokeai_config
-from .lora import LoRAModel, TextualInversionModel
 from .models import BaseModelType, ModelType, SubModelType, ModelBase
 
 # Maximum size of the cache, in gigs
@@ -187,7 +185,9 @@ class ModelCache(object):
         # TODO: lock for no copies on simultaneous calls?
         cache_entry = self._cached_models.get(key, None)
         if cache_entry is None:
-            self.logger.info(f"Loading model {model_path}, type {base_model}:{model_type}:{submodel}")
+            self.logger.info(
+                f"Loading model {model_path}, type {base_model.value}:{model_type.value}{':'+submodel.value if submodel else ''}"
+            )
 
             # this will remove older cached models until
             # there is sufficient room to load the requested model
@@ -358,7 +358,8 @@ class ModelCache(object):
             # 2 refs:
             # 1 from cache_entry
             # 1 from getrefcount function
-            if not cache_entry.locked and refs <= 2:
+            # 1 from onnx runtime object
+            if not cache_entry.locked and refs <= 3 if "onnx" in model_key else 2:
                 self.logger.debug(
                     f"Unloading model {model_key} to free {(model_size/GIG):.2f} GB (-{(cache_entry.size/GIG):.2f} GB)"
                 )

invokeai/backend/model_management/model_manager.py

@@ -228,19 +228,19 @@ the root is the InvokeAI ROOTDIR.
 """
 from __future__ import annotations
 
-import os
 import hashlib
+import os
 import textwrap
-import yaml
+import types
 from dataclasses import dataclass
 from pathlib import Path
-from typing import Optional, List, Tuple, Union, Dict, Set, Callable, types
 from shutil import rmtree, move
+from typing import Optional, List, Literal, Tuple, Union, Dict, Set, Callable
 
 import torch
+import yaml
 from omegaconf import OmegaConf
 from omegaconf.dictconfig import DictConfig
-
 from pydantic import BaseModel, Field
 
 import invokeai.backend.util.logging as logger
@@ -259,6 +259,7 @@ from .models import (
     ModelNotFoundException,
     InvalidModelException,
     DuplicateModelException,
+    ModelBase,
 )
 
 # We are only starting to number the config file with release 3.
@@ -276,7 +277,7 @@ class ModelInfo:
     hash: str
     location: Union[Path, str]
     precision: torch.dtype
-    _cache: ModelCache = None
+    _cache: Optional[ModelCache] = None
 
     def __enter__(self):
         return self.context.__enter__()
@@ -361,7 +362,7 @@ class ModelManager(object):
             if model_key.startswith("_"):
                 continue
             model_name, base_model, model_type = self.parse_key(model_key)
-            model_class = MODEL_CLASSES[base_model][model_type]
+            model_class = self._get_implementation(base_model, model_type)
             # alias for config file
             model_config["model_format"] = model_config.pop("format")
             self.models[model_key] = model_class.create_config(**model_config)
@@ -381,18 +382,24 @@ class ModelManager(object):
         # causing otherwise unreferenced models to be removed from memory
         self._read_models()
 
-    def model_exists(
-        self,
-        model_name: str,
-        base_model: BaseModelType,
-        model_type: ModelType,
-    ) -> bool:
+    def model_exists(self, model_name: str, base_model: BaseModelType, model_type: ModelType, *, rescan=False) -> bool:
         """
-        Given a model name, returns True if it is a valid
-        identifier.
+        Given a model name, returns True if it is a valid identifier.
+
+        :param model_name: symbolic name of the model in models.yaml
+        :param model_type: ModelType enum indicating the type of model to return
+        :param base_model: BaseModelType enum indicating the base model used by this model
+        :param rescan: if True, scan_models_directory
         """
         model_key = self.create_key(model_name, base_model, model_type)
-        return model_key in self.models
+        exists = model_key in self.models
+
+        # if model not found try to find it (maybe file just pasted)
+        if rescan and not exists:
+            self.scan_models_directory(base_model=base_model, model_type=model_type)
+            exists = self.model_exists(model_name, base_model, model_type, rescan=False)
+
+        return exists
 
     @classmethod
     def create_key(
@@ -423,7 +430,7 @@ class ModelManager(object):
         return (model_name, base_model, model_type)
 
     def _get_model_cache_path(self, model_path):
-        return self.app_config.models_path / ".cache" / hashlib.md5(str(model_path).encode()).hexdigest()
+        return self.resolve_model_path(Path(".cache") / hashlib.md5(str(model_path).encode()).hexdigest())
 
     @classmethod
     def initialize_model_config(cls, config_path: Path):
@@ -443,39 +450,32 @@ class ModelManager(object):
         :param model_name: symbolic name of the model in models.yaml
         :param model_type: ModelType enum indicating the type of model to return
         :param base_model: BaseModelType enum indicating the base model used by this model
-        :param submode_typel: an ModelType enum indicating the portion of
+        :param submodel_type: an ModelType enum indicating the portion of
                the model to retrieve (e.g. ModelType.Vae)
         """
-        model_class = MODEL_CLASSES[base_model][model_type]
         model_key = self.create_key(model_name, base_model, model_type)
 
-        # if model not found try to find it (maybe file just pasted)
-        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 ModelNotFoundException(f"Model not found - {model_key}")
+        if not self.model_exists(model_name, base_model, model_type, rescan=True):
+            raise ModelNotFoundException(f"Model not found - {model_key}")
 
-        model_config = self.models[model_key]
-        model_path = self.app_config.root_path / model_config.path
+        model_config = self._get_model_config(base_model, model_name, model_type)
+
+        model_path, is_submodel_override = self._get_model_path(model_config, submodel_type)
+
+        if is_submodel_override:
+            model_type = submodel_type
+            submodel_type = None
+
+        model_class = self._get_implementation(base_model, model_type)
 
         if not model_path.exists():
             if model_class.save_to_config:
                 self.models[model_key].error = ModelError.NotFound
-                raise Exception(f'Files for model "{model_key}" not found')
+                raise Exception(f'Files for model "{model_key}" not found at {model_path}')
 
             else:
                 self.models.pop(model_key, None)
-                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 = self.app_config.root_path / override_path
-                model_type = submodel_type
-                submodel_type = None
-                model_class = MODEL_CLASSES[base_model][model_type]
+                raise ModelNotFoundException(f'Files for model "{model_key}" not found at {model_path}')
 
         # TODO: path
         # TODO: is it accurate to use path as id
@@ -513,12 +513,61 @@ class ModelManager(object):
             _cache=self.cache,
         )
 
+    def _get_model_path(
+        self, model_config: ModelConfigBase, submodel_type: Optional[SubModelType] = None
+    ) -> (Path, bool):
+        """Extract a model's filesystem path from its config.
+
+        :return: The fully qualified Path of the module (or submodule).
+        """
+        model_path = model_config.path
+        is_submodel_override = False
+
+        # Does the config explicitly override the submodel?
+        if submodel_type is not None and hasattr(model_config, submodel_type):
+            submodel_path = getattr(model_config, submodel_type)
+            if submodel_path is not None:
+                model_path = getattr(model_config, submodel_type)
+                is_submodel_override = True
+
+        model_path = self.resolve_model_path(model_path)
+        return model_path, is_submodel_override
+
+    def _get_model_config(self, base_model: BaseModelType, model_name: str, model_type: ModelType) -> ModelConfigBase:
+        """Get a model's config object."""
+        model_key = self.create_key(model_name, base_model, model_type)
+        try:
+            model_config = self.models[model_key]
+        except KeyError:
+            raise ModelNotFoundException(f"Model not found - {model_key}")
+        return model_config
+
+    def _get_implementation(self, base_model: BaseModelType, model_type: ModelType) -> type[ModelBase]:
+        """Get the concrete implementation class for a specific model type."""
+        model_class = MODEL_CLASSES[base_model][model_type]
+        return model_class
+
+    def _instantiate(
+        self,
+        model_name: str,
+        base_model: BaseModelType,
+        model_type: ModelType,
+        submodel_type: Optional[SubModelType] = None,
+    ) -> ModelBase:
+        """Make a new instance of this model, without loading it."""
+        model_config = self._get_model_config(base_model, model_name, model_type)
+        model_path, is_submodel_override = self._get_model_path(model_config, submodel_type)
+        # FIXME: do non-overriden submodels get the right class?
+        constructor = self._get_implementation(base_model, model_type)
+        instance = constructor(model_path, base_model, model_type)
+        return instance
+
     def model_info(
         self,
         model_name: str,
         base_model: BaseModelType,
         model_type: ModelType,
-    ) -> dict:
+    ) -> Union[dict, None]:
         """
         Given a model name returns the OmegaConf (dict-like) object describing it.
         """
@@ -540,13 +589,15 @@ class ModelManager(object):
         model_name: str,
         base_model: BaseModelType,
         model_type: ModelType,
-    ) -> dict:
+    ) -> Union[dict, None]:
         """
         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
+        if len(models) > 0:
+            return models[0]
+        return None
 
     def list_models(
         self,
@@ -560,7 +611,7 @@ class ModelManager(object):
 
         model_keys = (
             [self.create_key(model_name, base_model, model_type)]
-            if model_name
+            if model_name and base_model and model_type
             else sorted(self.models, key=str.casefold)
         )
         models = []
@@ -586,7 +637,7 @@ class ModelManager(object):
 
             # expose paths as absolute to help web UI
             if path := model_dict.get("path"):
-                model_dict["path"] = str(self.app_config.root_path / path)
+                model_dict["path"] = str(self.resolve_model_path(path))
             models.append(model_dict)
 
         return models
@@ -596,7 +647,7 @@ class ModelManager(object):
         Print a table of models and their descriptions. This needs to be redone
         """
         # TODO: redo
-        for model_type, model_dict in self.list_models().items():
+        for model_dict in self.list_models():
             for model_name, model_info in model_dict.items():
                 line = f'{model_info["name"]:25s} {model_info["type"]:10s} {model_info["description"]}'
                 print(line)
@@ -623,7 +674,7 @@ class ModelManager(object):
             self.cache.uncache_model(cache_id)
 
         # if model inside invoke models folder - delete files
-        model_path = self.app_config.root_path / model_cfg.path
+        model_path = self.resolve_model_path(model_cfg.path)
         cache_path = self._get_model_cache_path(model_path)
         if cache_path.exists():
             rmtree(str(cache_path))
@@ -654,12 +705,11 @@ class ModelManager(object):
         The returned dict has the same format as the dict returned by
         model_info().
         """
-        # relativize paths as they go in - this makes it easier to move the root directory around
+        # relativize paths as they go in - this makes it easier to move the models directory around
         if path := model_attributes.get("path"):
-            if Path(path).is_relative_to(self.app_config.root_path):
-                model_attributes["path"] = str(Path(path).relative_to(self.app_config.root_path))
+            model_attributes["path"] = str(self.relative_model_path(Path(path)))
 
-        model_class = MODEL_CLASSES[base_model][model_type]
+        model_class = self._get_implementation(base_model, model_type)
         model_config = model_class.create_config(**model_attributes)
         model_key = self.create_key(model_name, base_model, model_type)
 
@@ -671,7 +721,7 @@ class ModelManager(object):
             # TODO: if path changed and old_model.path inside models folder should we delete this too?
 
             # remove conversion cache as config changed
-            old_model_path = self.app_config.root_path / old_model.path
+            old_model_path = self.resolve_model_path(old_model.path)
             old_model_cache = self._get_model_cache_path(old_model_path)
             if old_model_cache.exists():
                 if old_model_cache.is_dir():
@@ -700,8 +750,8 @@ class ModelManager(object):
         model_name: str,
         base_model: BaseModelType,
         model_type: ModelType,
-        new_name: str = None,
-        new_base: BaseModelType = None,
+        new_name: Optional[str] = None,
+        new_base: Optional[BaseModelType] = None,
     ):
         """
         Rename or rebase a model.
@@ -715,7 +765,7 @@ class ModelManager(object):
         if not model_cfg:
             raise ModelNotFoundException(f"Unknown model: {model_key}")
 
-        old_path = self.app_config.root_path / model_cfg.path
+        old_path = self.resolve_model_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)
@@ -724,15 +774,15 @@ class ModelManager(object):
 
         # 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"
-                / BaseModelType(new_base).value
-                / ModelType(model_type).value
-                / new_name
+            new_path = self.resolve_model_path(
+                Path(
+                    BaseModelType(new_base).value,
+                    ModelType(model_type).value,
+                    new_name,
+                )
             )
             move(old_path, new_path)
-            model_cfg.path = str(new_path.relative_to(self.app_config.root_path))
+            model_cfg.path = str(new_path.relative_to(self.app_config.models_path))
 
         # clean up caches
         old_model_cache = self._get_model_cache_path(old_path)
@@ -754,7 +804,7 @@ class ModelManager(object):
         self,
         model_name: str,
         base_model: BaseModelType,
-        model_type: Union[ModelType.Main, ModelType.Vae],
+        model_type: Literal[ModelType.Main, ModelType.Vae],
         dest_directory: Optional[Path] = None,
     ) -> AddModelResult:
         """
@@ -768,6 +818,10 @@ class ModelManager(object):
         This will raise a ValueError unless the model is a checkpoint.
         """
         info = self.model_info(model_name, base_model, model_type)
+
+        if info is None:
+            raise FileNotFoundError(f"model not found: {model_name}")
+
         if info["model_format"] != "checkpoint":
             raise ValueError(f"not a checkpoint format model: {model_name}")
 
@@ -781,8 +835,8 @@ class ModelManager(object):
             model_type,
             **submodel,
         )
-        checkpoint_path = self.app_config.root_path / info["path"]
-        old_diffusers_path = self.app_config.models_path / model.location
+        checkpoint_path = self.resolve_model_path(info["path"])
+        old_diffusers_path = self.resolve_model_path(model.location)
         new_diffusers_path = (
             dest_directory or self.app_config.models_path / base_model.value / model_type.value
         ) / model_name
@@ -795,7 +849,7 @@ class ModelManager(object):
             info["path"] = (
                 str(new_diffusers_path)
                 if dest_directory
-                else str(new_diffusers_path.relative_to(self.app_config.root_path))
+                else str(new_diffusers_path.relative_to(self.app_config.models_path))
             )
             info.pop("config")
 
@@ -810,6 +864,15 @@ class ModelManager(object):
 
         return result
 
+    def resolve_model_path(self, path: Union[Path, str]) -> Path:
+        """return relative paths based on configured models_path"""
+        return self.app_config.models_path / path
+
+    def relative_model_path(self, model_path: Path) -> Path:
+        if model_path.is_relative_to(self.app_config.models_path):
+            model_path = model_path.relative_to(self.app_config.models_path)
+        return model_path
+
     def search_models(self, search_folder):
         self.logger.info(f"Finding Models In: {search_folder}")
         models_folder_ckpt = Path(search_folder).glob("**/*.ckpt")
@@ -828,7 +891,7 @@ class ModelManager(object):
 
         return search_folder, found_models
 
-    def commit(self, conf_file: Path = None) -> None:
+    def commit(self, conf_file: Optional[Path] = None) -> None:
         """
         Write current configuration out to the indicated file.
         """
@@ -837,7 +900,7 @@ class ModelManager(object):
 
         for model_key, model_config in self.models.items():
             model_name, base_model, model_type = self.parse_key(model_key)
-            model_class = MODEL_CLASSES[base_model][model_type]
+            model_class = self._get_implementation(base_model, model_type)
             if model_class.save_to_config:
                 # TODO: or exclude_unset better fits here?
                 data_to_save[model_key] = model_config.dict(exclude_defaults=True, exclude={"error"})
@@ -883,12 +946,19 @@ class ModelManager(object):
         new_models_found = False
 
         self.logger.info(f"Scanning {self.app_config.models_path} for new models")
-        with Chdir(self.app_config.root_path):
+        with Chdir(self.app_config.models_path):
             for model_key, model_config in list(self.models.items()):
                 model_name, cur_base_model, cur_model_type = self.parse_key(model_key)
-                model_path = self.app_config.root_path.absolute() / model_config.path
+
+                # Patch for relative path bug in older models.yaml - paths should not
+                # be starting with a hard-coded 'models'. This will also fix up
+                # models.yaml when committed.
+                if model_config.path.startswith("models"):
+                    model_config.path = str(Path(*Path(model_config.path).parts[1:]))
+
+                model_path = self.resolve_model_path(model_config.path).absolute()
                 if not model_path.exists():
-                    model_class = MODEL_CLASSES[cur_base_model][cur_model_type]
+                    model_class = self._get_implementation(cur_base_model, cur_model_type)
                     if model_class.save_to_config:
                         model_config.error = ModelError.NotFound
                         self.models.pop(model_key, None)
@@ -904,8 +974,8 @@ class ModelManager(object):
                 for cur_model_type in ModelType:
                     if model_type is not None and cur_model_type != model_type:
                         continue
-                    model_class = MODEL_CLASSES[cur_base_model][cur_model_type]
-                    models_dir = self.app_config.models_path / cur_base_model.value / cur_model_type.value
+                    model_class = self._get_implementation(cur_base_model, cur_model_type)
+                    models_dir = self.resolve_model_path(Path(cur_base_model.value, cur_model_type.value))
 
                     if not models_dir.exists():
                         continue  # TODO: or create all folders?
@@ -919,9 +989,7 @@ class ModelManager(object):
                                 if model_key in self.models:
                                     raise DuplicateModelException(f"Model with key {model_key} added twice")
 
-                                if model_path.is_relative_to(self.app_config.root_path):
-                                    model_path = model_path.relative_to(self.app_config.root_path)
-
+                                model_path = self.relative_model_path(model_path)
                                 model_config: ModelConfigBase = model_class.probe_config(str(model_path))
                                 self.models[model_key] = model_config
                                 new_models_found = True
@@ -932,12 +1000,11 @@ class ModelManager(object):
                             except NotImplementedError as e:
                                 self.logger.warning(e)
 
-        imported_models = self.autoimport()
-
+        imported_models = self.scan_autoimport_directory()
         if (new_models_found or imported_models) and self.config_path:
             self.commit()
 
-    def autoimport(self) -> Dict[str, AddModelResult]:
+    def scan_autoimport_directory(self) -> Dict[str, AddModelResult]:
         """
         Scan the autoimport directory (if defined) and import new models, delete defunct models.
         """
@@ -971,7 +1038,7 @@ class ModelManager(object):
         # LS: hacky
         # Patch in the SD VAE from core so that it is available for use by the UI
         try:
-            self.heuristic_import({config.root_path / "models/core/convert/sd-vae-ft-mse"})
+            self.heuristic_import({str(self.resolve_model_path("core/convert/sd-vae-ft-mse"))})
         except:
             pass
 
@@ -980,7 +1047,7 @@ class ModelManager(object):
             model_manager=self,
             prediction_type_helper=ask_user_for_prediction_type,
         )
-        known_paths = {config.root_path / x["path"] for x in self.list_models()}
+        known_paths = {self.resolve_model_path(x["path"]) for x in self.list_models()}
         directories = {
             config.root_path / x
             for x in [
@@ -999,7 +1066,7 @@ class ModelManager(object):
     def heuristic_import(
         self,
         items_to_import: Set[str],
-        prediction_type_helper: Callable[[Path], SchedulerPredictionType] = None,
+        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.

invokeai/backend/model_management/model_merge.py

@@ -33,7 +33,7 @@ class ModelMerger(object):
         self,
         model_paths: List[Path],
         alpha: float = 0.5,
-        interp: MergeInterpolationMethod = None,
+        interp: Optional[MergeInterpolationMethod] = None,
         force: bool = False,
         **kwargs,
     ) -> DiffusionPipeline:
@@ -73,7 +73,7 @@ class ModelMerger(object):
         base_model: Union[BaseModelType, str],
         merged_model_name: str,
         alpha: float = 0.5,
-        interp: MergeInterpolationMethod = None,
+        interp: Optional[MergeInterpolationMethod] = None,
         force: bool = False,
         merge_dest_directory: Optional[Path] = None,
         **kwargs,
@@ -122,7 +122,7 @@ class ModelMerger(object):
         dump_path.mkdir(parents=True, exist_ok=True)
         dump_path = dump_path / merged_model_name
 
-        merged_pipe.save_pretrained(dump_path, safe_serialization=1)
+        merged_pipe.save_pretrained(dump_path, safe_serialization=True)
         attributes = dict(
             path=str(dump_path),
             description=f"Merge of models {', '.join(model_names)}",

invokeai/backend/model_management/model_probe.py

@@ -27,7 +27,7 @@ class ModelProbeInfo(object):
     variant_type: ModelVariantType
     prediction_type: SchedulerPredictionType
     upcast_attention: bool
-    format: Literal["diffusers", "checkpoint", "lycoris"]
+    format: Literal["diffusers", "checkpoint", "lycoris", "olive", "onnx"]
     image_size: int
 
 
@@ -41,6 +41,7 @@ class ModelProbe(object):
     PROBES = {
         "diffusers": {},
         "checkpoint": {},
+        "onnx": {},
     }
 
     CLASS2TYPE = {
@@ -53,7 +54,9 @@ class ModelProbe(object):
     }
 
     @classmethod
-    def register_probe(cls, format: Literal["diffusers", "checkpoint"], model_type: ModelType, probe_class: ProbeBase):
+    def register_probe(
+        cls, format: Literal["diffusers", "checkpoint", "onnx"], model_type: ModelType, probe_class: ProbeBase
+    ):
         cls.PROBES[format][model_type] = probe_class
 
     @classmethod
@@ -95,6 +98,7 @@ class ModelProbe(object):
                 if format_type == "diffusers"
                 else cls.get_model_type_from_checkpoint(model_path, model)
             )
+            format_type = "onnx" if model_type == ModelType.ONNX else format_type
             probe_class = cls.PROBES[format_type].get(model_type)
             if not probe_class:
                 return None
@@ -168,6 +172,8 @@ class ModelProbe(object):
         if model:
             class_name = model.__class__.__name__
         else:
+            if (folder_path / "unet/model.onnx").exists():
+                return ModelType.ONNX
             if (folder_path / "learned_embeds.bin").exists():
                 return ModelType.TextualInversion
 
@@ -309,21 +315,38 @@ class LoRACheckpointProbe(CheckpointProbeBase):
 
     def get_base_type(self) -> BaseModelType:
         checkpoint = self.checkpoint
+
+        # SD-2 models are very hard to probe. These probes are brittle and likely to fail in the future
+        # There are also some "SD-2 LoRAs" that have identical keys and shapes to SD-1 and will be
+        # misclassified as SD-1
+        key = "lora_te_text_model_encoder_layers_0_mlp_fc1.lora_down.weight"
+        if key in checkpoint and checkpoint[key].shape[0] == 320:
+            return BaseModelType.StableDiffusion2
+
+        key = "lora_unet_output_blocks_5_1_transformer_blocks_1_ff_net_2.lora_up.weight"
+        if key in checkpoint:
+            return BaseModelType.StableDiffusionXL
+
         key1 = "lora_te_text_model_encoder_layers_0_mlp_fc1.lora_down.weight"
-        key2 = "lora_te_text_model_encoder_layers_0_self_attn_k_proj.hada_w1_a"
+        key2 = "lora_te_text_model_encoder_layers_0_self_attn_k_proj.lora_down.weight"
+        key3 = "lora_te_text_model_encoder_layers_0_self_attn_k_proj.hada_w1_a"
+
         lora_token_vector_length = (
             checkpoint[key1].shape[1]
             if key1 in checkpoint
-            else checkpoint[key2].shape[0]
+            else checkpoint[key2].shape[1]
             if key2 in checkpoint
-            else 768
+            else checkpoint[key3].shape[0]
+            if key3 in checkpoint
+            else None
         )
+
         if lora_token_vector_length == 768:
             return BaseModelType.StableDiffusion1
         elif lora_token_vector_length == 1024:
             return BaseModelType.StableDiffusion2
         else:
-            return None
+            raise InvalidModelException(f"Unknown LoRA type")
 
 
 class TextualInversionCheckpointProbe(CheckpointProbeBase):
@@ -460,6 +483,17 @@ class TextualInversionFolderProbe(FolderProbeBase):
         return TextualInversionCheckpointProbe(None, checkpoint=checkpoint).get_base_type()
 
 
+class ONNXFolderProbe(FolderProbeBase):
+    def get_format(self) -> str:
+        return "onnx"
+
+    def get_base_type(self) -> BaseModelType:
+        return BaseModelType.StableDiffusion1
+
+    def get_variant_type(self) -> ModelVariantType:
+        return ModelVariantType.Normal
+
+
 class ControlNetFolderProbe(FolderProbeBase):
     def get_base_type(self) -> BaseModelType:
         config_file = self.folder_path / "config.json"
@@ -497,3 +531,4 @@ ModelProbe.register_probe("checkpoint", ModelType.Vae, VaeCheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.Lora, LoRACheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.TextualInversion, TextualInversionCheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.ControlNet, ControlNetCheckpointProbe)
+ModelProbe.register_probe("onnx", ModelType.ONNX, ONNXFolderProbe)

invokeai/backend/model_management/models/__init__.py

@@ -23,8 +23,11 @@ from .lora import LoRAModel
 from .controlnet import ControlNetModel  # TODO:
 from .textual_inversion import TextualInversionModel
 
+from .stable_diffusion_onnx import ONNXStableDiffusion1Model, ONNXStableDiffusion2Model
+
 MODEL_CLASSES = {
     BaseModelType.StableDiffusion1: {
+        ModelType.ONNX: ONNXStableDiffusion1Model,
         ModelType.Main: StableDiffusion1Model,
         ModelType.Vae: VaeModel,
         ModelType.Lora: LoRAModel,
@@ -32,6 +35,7 @@ MODEL_CLASSES = {
         ModelType.TextualInversion: TextualInversionModel,
     },
     BaseModelType.StableDiffusion2: {
+        ModelType.ONNX: ONNXStableDiffusion2Model,
         ModelType.Main: StableDiffusion2Model,
         ModelType.Vae: VaeModel,
         ModelType.Lora: LoRAModel,
@@ -45,6 +49,7 @@ MODEL_CLASSES = {
         ModelType.Lora: LoRAModel,
         ModelType.ControlNet: ControlNetModel,
         ModelType.TextualInversion: TextualInversionModel,
+        ModelType.ONNX: ONNXStableDiffusion2Model,
     },
     BaseModelType.StableDiffusionXLRefiner: {
         ModelType.Main: StableDiffusionXLModel,
@@ -53,6 +58,7 @@ MODEL_CLASSES = {
         ModelType.Lora: LoRAModel,
         ModelType.ControlNet: ControlNetModel,
         ModelType.TextualInversion: TextualInversionModel,
+        ModelType.ONNX: ONNXStableDiffusion2Model,
     },
     # BaseModelType.Kandinsky2_1: {
     #    ModelType.Main: Kandinsky2_1Model,

invokeai/backend/model_management/models/base.py

@@ -8,13 +8,23 @@ from abc import ABCMeta, abstractmethod
 from pathlib import Path
 from picklescan.scanner import scan_file_path
 import torch
+import numpy as np
 import safetensors.torch
-from diffusers import DiffusionPipeline, ConfigMixin
+from pathlib import Path
+from diffusers import DiffusionPipeline, ConfigMixin, OnnxRuntimeModel
 
 from contextlib import suppress
 from pydantic import BaseModel, Field
 from typing import List, Dict, Optional, Type, Literal, TypeVar, Generic, Callable, Any, Union
 
+import onnx
+from onnx import numpy_helper
+from onnxruntime import (
+    InferenceSession,
+    SessionOptions,
+    get_available_providers,
+)
+
 
 class DuplicateModelException(Exception):
     pass
@@ -37,6 +47,7 @@ class BaseModelType(str, Enum):
 
 
 class ModelType(str, Enum):
+    ONNX = "onnx"
     Main = "main"
     Vae = "vae"
     Lora = "lora"
@@ -51,6 +62,8 @@ class SubModelType(str, Enum):
     Tokenizer = "tokenizer"
     Tokenizer2 = "tokenizer_2"
     Vae = "vae"
+    VaeDecoder = "vae_decoder"
+    VaeEncoder = "vae_encoder"
     Scheduler = "scheduler"
     SafetyChecker = "safety_checker"
     # MoVQ = "movq"
@@ -279,8 +292,9 @@ class DiffusersModel(ModelBase):
                 )
                 break
             except Exception as e:
-                # print("====ERR LOAD====")
-                # print(f"{variant}: {e}")
+                if not str(e).startswith("Error no file"):
+                    print("====ERR LOAD====")
+                    print(f"{variant}: {e}")
                 pass
         else:
             raise Exception(f"Failed to load {self.base_model}:{self.model_type}:{child_type} model")
@@ -362,6 +376,8 @@ def calc_model_size_by_data(model) -> int:
         return _calc_pipeline_by_data(model)
     elif isinstance(model, torch.nn.Module):
         return _calc_model_by_data(model)
+    elif isinstance(model, IAIOnnxRuntimeModel):
+        return _calc_onnx_model_by_data(model)
     else:
         return 0
 
@@ -382,6 +398,12 @@ def _calc_model_by_data(model) -> int:
     return mem
 
 
+def _calc_onnx_model_by_data(model) -> int:
+    tensor_size = model.tensors.size() * 2  # The session doubles this
+    mem = tensor_size  # in bytes
+    return mem
+
+
 def _fast_safetensors_reader(path: str):
     checkpoint = dict()
     device = torch.device("meta")
@@ -449,3 +471,208 @@ class SilenceWarnings(object):
         transformers_logging.set_verbosity(self.transformers_verbosity)
         diffusers_logging.set_verbosity(self.diffusers_verbosity)
         warnings.simplefilter("default")
+
+
+ONNX_WEIGHTS_NAME = "model.onnx"
+
+
+class IAIOnnxRuntimeModel:
+    class _tensor_access:
+        def __init__(self, model):
+            self.model = model
+            self.indexes = dict()
+            for idx, obj in enumerate(self.model.proto.graph.initializer):
+                self.indexes[obj.name] = idx
+
+        def __getitem__(self, key: str):
+            value = self.model.proto.graph.initializer[self.indexes[key]]
+            return numpy_helper.to_array(value)
+
+        def __setitem__(self, key: str, value: np.ndarray):
+            new_node = numpy_helper.from_array(value)
+            # set_external_data(new_node, location="in-memory-location")
+            new_node.name = key
+            # new_node.ClearField("raw_data")
+            del self.model.proto.graph.initializer[self.indexes[key]]
+            self.model.proto.graph.initializer.insert(self.indexes[key], new_node)
+            # self.model.data[key] = OrtValue.ortvalue_from_numpy(value)
+
+        # __delitem__
+
+        def __contains__(self, key: str):
+            return self.indexes[key] in self.model.proto.graph.initializer
+
+        def items(self):
+            raise NotImplementedError("tensor.items")
+            # return [(obj.name, obj) for obj in self.raw_proto]
+
+        def keys(self):
+            return self.indexes.keys()
+
+        def values(self):
+            raise NotImplementedError("tensor.values")
+            # return [obj for obj in self.raw_proto]
+
+        def size(self):
+            bytesSum = 0
+            for node in self.model.proto.graph.initializer:
+                bytesSum += sys.getsizeof(node.raw_data)
+            return bytesSum
+
+    class _access_helper:
+        def __init__(self, raw_proto):
+            self.indexes = dict()
+            self.raw_proto = raw_proto
+            for idx, obj in enumerate(raw_proto):
+                self.indexes[obj.name] = idx
+
+        def __getitem__(self, key: str):
+            return self.raw_proto[self.indexes[key]]
+
+        def __setitem__(self, key: str, value):
+            index = self.indexes[key]
+            del self.raw_proto[index]
+            self.raw_proto.insert(index, value)
+
+        # __delitem__
+
+        def __contains__(self, key: str):
+            return key in self.indexes
+
+        def items(self):
+            return [(obj.name, obj) for obj in self.raw_proto]
+
+        def keys(self):
+            return self.indexes.keys()
+
+        def values(self):
+            return [obj for obj in self.raw_proto]
+
+    def __init__(self, model_path: str, provider: Optional[str]):
+        self.path = model_path
+        self.session = None
+        self.provider = provider
+        """
+        self.data_path = self.path + "_data"
+        if not os.path.exists(self.data_path):
+            print(f"Moving model tensors to separate file: {self.data_path}")
+            tmp_proto = onnx.load(model_path, load_external_data=True)
+            onnx.save_model(tmp_proto, self.path, save_as_external_data=True, all_tensors_to_one_file=True, location=os.path.basename(self.data_path), size_threshold=1024, convert_attribute=False)
+            del tmp_proto
+            gc.collect()
+
+        self.proto = onnx.load(model_path, load_external_data=False)
+        """
+
+        self.proto = onnx.load(model_path, load_external_data=True)
+        # self.data = dict()
+        # for tensor in self.proto.graph.initializer:
+        #     name = tensor.name
+
+        #     if tensor.HasField("raw_data"):
+        #         npt = numpy_helper.to_array(tensor)
+        #         orv = OrtValue.ortvalue_from_numpy(npt)
+        #         # self.data[name] = orv
+        #         # set_external_data(tensor, location="in-memory-location")
+        #         tensor.name = name
+        #         # tensor.ClearField("raw_data")
+
+        self.nodes = self._access_helper(self.proto.graph.node)
+        # self.initializers = self._access_helper(self.proto.graph.initializer)
+        # print(self.proto.graph.input)
+        # print(self.proto.graph.initializer)
+
+        self.tensors = self._tensor_access(self)
+
+    # TODO: integrate with model manager/cache
+    def create_session(self, height=None, width=None):
+        if self.session is None or self.session_width != width or self.session_height != height:
+            # onnx.save(self.proto, "tmp.onnx")
+            # onnx.save_model(self.proto, "tmp.onnx", save_as_external_data=True, all_tensors_to_one_file=True, location="tmp.onnx_data", size_threshold=1024, convert_attribute=False)
+            # TODO: something to be able to get weight when they already moved outside of model proto
+            # (trimmed_model, external_data) = buffer_external_data_tensors(self.proto)
+            sess = SessionOptions()
+            # self._external_data.update(**external_data)
+            # sess.add_external_initializers(list(self.data.keys()), list(self.data.values()))
+            # sess.enable_profiling = True
+
+            # sess.intra_op_num_threads = 1
+            # sess.inter_op_num_threads = 1
+            # sess.execution_mode = ExecutionMode.ORT_SEQUENTIAL
+            # sess.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL
+            # sess.enable_cpu_mem_arena = True
+            # sess.enable_mem_pattern = True
+            # sess.add_session_config_entry("session.intra_op.use_xnnpack_threadpool", "1") ########### It's the key code
+            self.session_height = height
+            self.session_width = width
+            if height and width:
+                sess.add_free_dimension_override_by_name("unet_sample_batch", 2)
+                sess.add_free_dimension_override_by_name("unet_sample_channels", 4)
+                sess.add_free_dimension_override_by_name("unet_hidden_batch", 2)
+                sess.add_free_dimension_override_by_name("unet_hidden_sequence", 77)
+                sess.add_free_dimension_override_by_name("unet_sample_height", self.session_height)
+                sess.add_free_dimension_override_by_name("unet_sample_width", self.session_width)
+                sess.add_free_dimension_override_by_name("unet_time_batch", 1)
+            providers = []
+            if self.provider:
+                providers.append(self.provider)
+            else:
+                providers = get_available_providers()
+            if "TensorrtExecutionProvider" in providers:
+                providers.remove("TensorrtExecutionProvider")
+            try:
+                self.session = InferenceSession(self.proto.SerializeToString(), providers=providers, sess_options=sess)
+            except Exception as e:
+                raise e
+            # self.session = InferenceSession("tmp.onnx", providers=[self.provider], sess_options=self.sess_options)
+            # self.io_binding = self.session.io_binding()
+
+    def release_session(self):
+        self.session = None
+        import gc
+
+        gc.collect()
+        return
+
+    def __call__(self, **kwargs):
+        if self.session is None:
+            raise Exception("You should call create_session before running model")
+
+        inputs = {k: np.array(v) for k, v in kwargs.items()}
+        output_names = self.session.get_outputs()
+        # for k in inputs:
+        #     self.io_binding.bind_cpu_input(k, inputs[k])
+        # for name in output_names:
+        #     self.io_binding.bind_output(name.name)
+        # self.session.run_with_iobinding(self.io_binding, None)
+        # return self.io_binding.copy_outputs_to_cpu()
+        return self.session.run(None, inputs)
+
+    # compatability with diffusers load code
+    @classmethod
+    def from_pretrained(
+        cls,
+        model_id: Union[str, Path],
+        subfolder: Union[str, Path] = None,
+        file_name: Optional[str] = None,
+        provider: Optional[str] = None,
+        sess_options: Optional["SessionOptions"] = None,
+        **kwargs,
+    ):
+        file_name = file_name or ONNX_WEIGHTS_NAME
+
+        if os.path.isdir(model_id):
+            model_path = model_id
+            if subfolder is not None:
+                model_path = os.path.join(model_path, subfolder)
+            model_path = os.path.join(model_path, file_name)
+
+        else:
+            model_path = model_id
+
+        # load model from local directory
+        if not os.path.isfile(model_path):
+            raise Exception(f"Model not found: {model_path}")
+
+        # TODO: session options
+        return cls(model_path, provider=provider)

invokeai/backend/model_management/models/controlnet.py

@@ -17,6 +17,7 @@ from .base import (
     ModelNotFoundException,
 )
 from invokeai.app.services.config import InvokeAIAppConfig
+import invokeai.backend.util.logging as logger
 
 
 class ControlNetModelFormat(str, Enum):
@@ -66,7 +67,7 @@ class ControlNetModel(ModelBase):
         child_type: Optional[SubModelType] = None,
     ):
         if child_type is not None:
-            raise Exception("There is no child models in controlnet model")
+            raise Exception("There are no child models in controlnet model")
 
         model = None
         for variant in ["fp16", None]:
@@ -124,9 +125,7 @@ class ControlNetModel(ModelBase):
             return model_path
 
 
-@classmethod
 def _convert_controlnet_ckpt_and_cache(
-    cls,
     model_path: str,
     output_path: str,
     base_model: BaseModelType,
@@ -141,6 +140,7 @@ def _convert_controlnet_ckpt_and_cache(
     weights = app_config.root_path / model_path
     output_path = Path(output_path)
 
+    logger.info(f"Converting {weights} to diffusers format")
     # return cached version if it exists
     if output_path.exists():
         return output_path

invokeai/backend/model_management/models/lora.py

@@ -1,7 +1,9 @@
 import os
 import torch
 from enum import Enum
-from typing import Optional, Union, Literal
+from typing import Optional, Dict, Union, Literal, Any
+from pathlib import Path
+from safetensors.torch import load_file
 from .base import (
     ModelBase,
     ModelConfigBase,
@@ -13,9 +15,6 @@ from .base import (
     ModelNotFoundException,
 )
 
-# TODO: naming
-from ..lora import LoRAModel as LoRAModelRaw
-
 
 class LoRAModelFormat(str, Enum):
     LyCORIS = "lycoris"
@@ -50,6 +49,7 @@ class LoRAModel(ModelBase):
         model = LoRAModelRaw.from_checkpoint(
             file_path=self.model_path,
             dtype=torch_dtype,
+            base_model=self.base_model,
         )
 
         self.model_size = model.calc_size()
@@ -87,3 +87,582 @@ class LoRAModel(ModelBase):
             raise NotImplementedError("Diffusers lora not supported")
         else:
             return model_path
+
+
+class LoRALayerBase:
+    # rank: Optional[int]
+    # alpha: Optional[float]
+    # bias: Optional[torch.Tensor]
+    # layer_key: str
+
+    # @property
+    # def scale(self):
+    #    return self.alpha / self.rank if (self.alpha and self.rank) else 1.0
+
+    def __init__(
+        self,
+        layer_key: str,
+        values: dict,
+    ):
+        if "alpha" in values:
+            self.alpha = values["alpha"].item()
+        else:
+            self.alpha = None
+
+        if "bias_indices" in values and "bias_values" in values and "bias_size" in values:
+            self.bias = torch.sparse_coo_tensor(
+                values["bias_indices"],
+                values["bias_values"],
+                tuple(values["bias_size"]),
+            )
+
+        else:
+            self.bias = None
+
+        self.rank = None  # set in layer implementation
+        self.layer_key = layer_key
+
+    def forward(
+        self,
+        module: torch.nn.Module,
+        input_h: Any,  # for real looks like Tuple[torch.nn.Tensor] but not sure
+        multiplier: float,
+    ):
+        if type(module) == torch.nn.Conv2d:
+            op = torch.nn.functional.conv2d
+            extra_args = dict(
+                stride=module.stride,
+                padding=module.padding,
+                dilation=module.dilation,
+                groups=module.groups,
+            )
+
+        else:
+            op = torch.nn.functional.linear
+            extra_args = {}
+
+        weight = self.get_weight()
+
+        bias = self.bias if self.bias is not None else 0
+        scale = self.alpha / self.rank if (self.alpha and self.rank) else 1.0
+        return (
+            op(
+                *input_h,
+                (weight + bias).view(module.weight.shape),
+                None,
+                **extra_args,
+            )
+            * multiplier
+            * scale
+        )
+
+    def get_weight(self):
+        raise NotImplementedError()
+
+    def calc_size(self) -> int:
+        model_size = 0
+        for val in [self.bias]:
+            if val is not None:
+                model_size += val.nelement() * val.element_size()
+        return model_size
+
+    def to(
+        self,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        if self.bias is not None:
+            self.bias = self.bias.to(device=device, dtype=dtype)
+
+
+# TODO: find and debug lora/locon with bias
+class LoRALayer(LoRALayerBase):
+    # up: torch.Tensor
+    # mid: Optional[torch.Tensor]
+    # down: torch.Tensor
+
+    def __init__(
+        self,
+        layer_key: str,
+        values: dict,
+    ):
+        super().__init__(layer_key, values)
+
+        self.up = values["lora_up.weight"]
+        self.down = values["lora_down.weight"]
+        if "lora_mid.weight" in values:
+            self.mid = values["lora_mid.weight"]
+        else:
+            self.mid = None
+
+        self.rank = self.down.shape[0]
+
+    def get_weight(self):
+        if self.mid is not None:
+            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)
+
+        return weight
+
+    def calc_size(self) -> int:
+        model_size = super().calc_size()
+        for val in [self.up, self.mid, self.down]:
+            if val is not None:
+                model_size += val.nelement() * val.element_size()
+        return model_size
+
+    def to(
+        self,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        super().to(device=device, dtype=dtype)
+
+        self.up = self.up.to(device=device, dtype=dtype)
+        self.down = self.down.to(device=device, dtype=dtype)
+
+        if self.mid is not None:
+            self.mid = self.mid.to(device=device, dtype=dtype)
+
+
+class LoHALayer(LoRALayerBase):
+    # w1_a: torch.Tensor
+    # w1_b: torch.Tensor
+    # w2_a: torch.Tensor
+    # w2_b: torch.Tensor
+    # t1: Optional[torch.Tensor] = None
+    # t2: Optional[torch.Tensor] = None
+
+    def __init__(
+        self,
+        layer_key: str,
+        values: dict,
+    ):
+        super().__init__(layer_key, values)
+
+        self.w1_a = values["hada_w1_a"]
+        self.w1_b = values["hada_w1_b"]
+        self.w2_a = values["hada_w2_a"]
+        self.w2_b = values["hada_w2_b"]
+
+        if "hada_t1" in values:
+            self.t1 = values["hada_t1"]
+        else:
+            self.t1 = None
+
+        if "hada_t2" in values:
+            self.t2 = values["hada_t2"]
+        else:
+            self.t2 = None
+
+        self.rank = self.w1_b.shape[0]
+
+    def get_weight(self):
+        if self.t1 is None:
+            weight = (self.w1_a @ self.w1_b) * (self.w2_a @ self.w2_b)
+
+        else:
+            rebuild1 = torch.einsum("i j k l, j r, i p -> p r k l", self.t1, self.w1_b, self.w1_a)
+            rebuild2 = torch.einsum("i j k l, j r, i p -> p r k l", self.t2, self.w2_b, self.w2_a)
+            weight = rebuild1 * rebuild2
+
+        return weight
+
+    def calc_size(self) -> int:
+        model_size = super().calc_size()
+        for val in [self.w1_a, self.w1_b, self.w2_a, self.w2_b, self.t1, self.t2]:
+            if val is not None:
+                model_size += val.nelement() * val.element_size()
+        return model_size
+
+    def to(
+        self,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        super().to(device=device, dtype=dtype)
+
+        self.w1_a = self.w1_a.to(device=device, dtype=dtype)
+        self.w1_b = self.w1_b.to(device=device, dtype=dtype)
+        if self.t1 is not None:
+            self.t1 = self.t1.to(device=device, dtype=dtype)
+
+        self.w2_a = self.w2_a.to(device=device, dtype=dtype)
+        self.w2_b = self.w2_b.to(device=device, dtype=dtype)
+        if self.t2 is not None:
+            self.t2 = self.t2.to(device=device, dtype=dtype)
+
+
+class LoKRLayer(LoRALayerBase):
+    # w1: Optional[torch.Tensor] = None
+    # w1_a: Optional[torch.Tensor] = None
+    # w1_b: Optional[torch.Tensor] = None
+    # w2: Optional[torch.Tensor] = None
+    # w2_a: Optional[torch.Tensor] = None
+    # w2_b: Optional[torch.Tensor] = None
+    # t2: Optional[torch.Tensor] = None
+
+    def __init__(
+        self,
+        layer_key: str,
+        values: dict,
+    ):
+        super().__init__(layer_key, values)
+
+        if "lokr_w1" in values:
+            self.w1 = values["lokr_w1"]
+            self.w1_a = None
+            self.w1_b = None
+        else:
+            self.w1 = None
+            self.w1_a = values["lokr_w1_a"]
+            self.w1_b = values["lokr_w1_b"]
+
+        if "lokr_w2" in values:
+            self.w2 = values["lokr_w2"]
+            self.w2_a = None
+            self.w2_b = None
+        else:
+            self.w2 = None
+            self.w2_a = values["lokr_w2_a"]
+            self.w2_b = values["lokr_w2_b"]
+
+        if "lokr_t2" in values:
+            self.t2 = values["lokr_t2"]
+        else:
+            self.t2 = None
+
+        if "lokr_w1_b" in values:
+            self.rank = values["lokr_w1_b"].shape[0]
+        elif "lokr_w2_b" in values:
+            self.rank = values["lokr_w2_b"].shape[0]
+        else:
+            self.rank = None  # unscaled
+
+    def get_weight(self):
+        w1 = self.w1
+        if w1 is None:
+            w1 = self.w1_a @ self.w1_b
+
+        w2 = self.w2
+        if w2 is None:
+            if self.t2 is None:
+                w2 = self.w2_a @ self.w2_b
+            else:
+                w2 = torch.einsum("i j k l, i p, j r -> p r k l", self.t2, self.w2_a, self.w2_b)
+
+        if len(w2.shape) == 4:
+            w1 = w1.unsqueeze(2).unsqueeze(2)
+        w2 = w2.contiguous()
+        weight = torch.kron(w1, w2)
+
+        return weight
+
+    def calc_size(self) -> int:
+        model_size = super().calc_size()
+        for val in [self.w1, self.w1_a, self.w1_b, self.w2, self.w2_a, self.w2_b, self.t2]:
+            if val is not None:
+                model_size += val.nelement() * val.element_size()
+        return model_size
+
+    def to(
+        self,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        super().to(device=device, dtype=dtype)
+
+        if self.w1 is not None:
+            self.w1 = self.w1.to(device=device, dtype=dtype)
+        else:
+            self.w1_a = self.w1_a.to(device=device, dtype=dtype)
+            self.w1_b = self.w1_b.to(device=device, dtype=dtype)
+
+        if self.w2 is not None:
+            self.w2 = self.w2.to(device=device, dtype=dtype)
+        else:
+            self.w2_a = self.w2_a.to(device=device, dtype=dtype)
+            self.w2_b = self.w2_b.to(device=device, dtype=dtype)
+
+        if self.t2 is not None:
+            self.t2 = self.t2.to(device=device, dtype=dtype)
+
+
+class FullLayer(LoRALayerBase):
+    # weight: torch.Tensor
+
+    def __init__(
+        self,
+        layer_key: str,
+        values: dict,
+    ):
+        super().__init__(layer_key, values)
+
+        self.weight = values["diff"]
+
+        if len(values.keys()) > 1:
+            _keys = list(values.keys())
+            _keys.remove("diff")
+            raise NotImplementedError(f"Unexpected keys in lora diff layer: {_keys}")
+
+        self.rank = None  # unscaled
+
+    def get_weight(self):
+        return self.weight
+
+    def calc_size(self) -> int:
+        model_size = super().calc_size()
+        model_size += self.weight.nelement() * self.weight.element_size()
+        return model_size
+
+    def to(
+        self,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        super().to(device=device, dtype=dtype)
+
+        self.weight = self.weight.to(device=device, dtype=dtype)
+
+
+# TODO: rename all methods used in model logic with Info postfix and remove here Raw postfix
+class LoRAModelRaw:  # (torch.nn.Module):
+    _name: str
+    layers: Dict[str, LoRALayer]
+    _device: torch.device
+    _dtype: torch.dtype
+
+    def __init__(
+        self,
+        name: str,
+        layers: Dict[str, LoRALayer],
+        device: torch.device,
+        dtype: torch.dtype,
+    ):
+        self._name = name
+        self._device = device or torch.cpu
+        self._dtype = dtype or torch.float32
+        self.layers = layers
+
+    @property
+    def name(self):
+        return self._name
+
+    @property
+    def device(self):
+        return self._device
+
+    @property
+    def dtype(self):
+        return self._dtype
+
+    def to(
+        self,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        # TODO: try revert if exception?
+        for key, layer in self.layers.items():
+            layer.to(device=device, dtype=dtype)
+        self._device = device
+        self._dtype = dtype
+
+    def calc_size(self) -> int:
+        model_size = 0
+        for _, layer in self.layers.items():
+            model_size += layer.calc_size()
+        return model_size
+
+    @classmethod
+    def _convert_sdxl_compvis_keys(cls, state_dict):
+        new_state_dict = dict()
+        for full_key, value in state_dict.items():
+            if full_key.startswith("lora_te1_") or full_key.startswith("lora_te2_"):
+                continue  # clip same
+
+            if not full_key.startswith("lora_unet_"):
+                raise NotImplementedError(f"Unknown prefix for sdxl lora key - {full_key}")
+            src_key = full_key.replace("lora_unet_", "")
+            try:
+                dst_key = None
+                while "_" in src_key:
+                    if src_key in SDXL_UNET_COMPVIS_MAP:
+                        dst_key = SDXL_UNET_COMPVIS_MAP[src_key]
+                        break
+                    src_key = "_".join(src_key.split("_")[:-1])
+
+                if dst_key is None:
+                    raise Exception(f"Unknown sdxl lora key - {full_key}")
+                new_key = full_key.replace(src_key, dst_key)
+            except:
+                print(SDXL_UNET_COMPVIS_MAP)
+                raise
+            new_state_dict[new_key] = value
+        return new_state_dict
+
+    @classmethod
+    def from_checkpoint(
+        cls,
+        file_path: Union[str, Path],
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        base_model: Optional[BaseModelType] = None,
+    ):
+        device = device or torch.device("cpu")
+        dtype = dtype or torch.float32
+
+        if isinstance(file_path, str):
+            file_path = Path(file_path)
+
+        model = cls(
+            device=device,
+            dtype=dtype,
+            name=file_path.stem,  # TODO:
+            layers=dict(),
+        )
+
+        if file_path.suffix == ".safetensors":
+            state_dict = load_file(file_path.absolute().as_posix(), device="cpu")
+        else:
+            state_dict = torch.load(file_path, map_location="cpu")
+
+        state_dict = cls._group_state(state_dict)
+
+        if base_model == BaseModelType.StableDiffusionXL:
+            state_dict = cls._convert_sdxl_compvis_keys(state_dict)
+
+        for layer_key, values in state_dict.items():
+            # lora and locon
+            if "lora_down.weight" in values:
+                layer = LoRALayer(layer_key, values)
+
+            # loha
+            elif "hada_w1_b" in values:
+                layer = LoHALayer(layer_key, values)
+
+            # lokr
+            elif "lokr_w1_b" in values or "lokr_w1" in values:
+                layer = LoKRLayer(layer_key, values)
+
+            elif "diff" in values:
+                layer = FullLayer(layer_key, values)
+
+            else:
+                # TODO: ia3/... format
+                print(f">> Encountered unknown lora layer module in {model.name}: {layer_key} - {list(values.keys())}")
+                raise Exception("Unknown lora format!")
+
+            # lower memory consumption by removing already parsed layer values
+            state_dict[layer_key].clear()
+
+            layer.to(device=device, dtype=dtype)
+            model.layers[layer_key] = layer
+
+        return model
+
+    @staticmethod
+    def _group_state(state_dict: dict):
+        state_dict_groupped = dict()
+
+        for key, value in state_dict.items():
+            stem, leaf = key.split(".", 1)
+            if stem not in state_dict_groupped:
+                state_dict_groupped[stem] = dict()
+            state_dict_groupped[stem][leaf] = value
+
+        return state_dict_groupped
+
+
+# code from
+# https://github.com/bmaltais/kohya_ss/blob/2accb1305979ba62f5077a23aabac23b4c37e935/networks/lora_diffusers.py#L15C1-L97C32
+def make_sdxl_unet_conversion_map():
+    unet_conversion_map_layer = []
+
+    for i in range(3):  # num_blocks is 3 in sdxl
+        # loop over downblocks/upblocks
+        for j in range(2):
+            # loop over resnets/attentions for downblocks
+            hf_down_res_prefix = f"down_blocks.{i}.resnets.{j}."
+            sd_down_res_prefix = f"input_blocks.{3*i + j + 1}.0."
+            unet_conversion_map_layer.append((sd_down_res_prefix, hf_down_res_prefix))
+
+            if i < 3:
+                # no attention layers in down_blocks.3
+                hf_down_atn_prefix = f"down_blocks.{i}.attentions.{j}."
+                sd_down_atn_prefix = f"input_blocks.{3*i + j + 1}.1."
+                unet_conversion_map_layer.append((sd_down_atn_prefix, hf_down_atn_prefix))
+
+        for j in range(3):
+            # loop over resnets/attentions for upblocks
+            hf_up_res_prefix = f"up_blocks.{i}.resnets.{j}."
+            sd_up_res_prefix = f"output_blocks.{3*i + j}.0."
+            unet_conversion_map_layer.append((sd_up_res_prefix, hf_up_res_prefix))
+
+            # if i > 0: commentout for sdxl
+            # no attention layers in up_blocks.0
+            hf_up_atn_prefix = f"up_blocks.{i}.attentions.{j}."
+            sd_up_atn_prefix = f"output_blocks.{3*i + j}.1."
+            unet_conversion_map_layer.append((sd_up_atn_prefix, hf_up_atn_prefix))
+
+        if i < 3:
+            # no downsample in down_blocks.3
+            hf_downsample_prefix = f"down_blocks.{i}.downsamplers.0.conv."
+            sd_downsample_prefix = f"input_blocks.{3*(i+1)}.0.op."
+            unet_conversion_map_layer.append((sd_downsample_prefix, hf_downsample_prefix))
+
+            # no upsample in up_blocks.3
+            hf_upsample_prefix = f"up_blocks.{i}.upsamplers.0."
+            sd_upsample_prefix = f"output_blocks.{3*i + 2}.{2}."  # change for sdxl
+            unet_conversion_map_layer.append((sd_upsample_prefix, hf_upsample_prefix))
+
+    hf_mid_atn_prefix = "mid_block.attentions.0."
+    sd_mid_atn_prefix = "middle_block.1."
+    unet_conversion_map_layer.append((sd_mid_atn_prefix, hf_mid_atn_prefix))
+
+    for j in range(2):
+        hf_mid_res_prefix = f"mid_block.resnets.{j}."
+        sd_mid_res_prefix = f"middle_block.{2*j}."
+        unet_conversion_map_layer.append((sd_mid_res_prefix, hf_mid_res_prefix))
+
+    unet_conversion_map_resnet = [
+        # (stable-diffusion, HF Diffusers)
+        ("in_layers.0.", "norm1."),
+        ("in_layers.2.", "conv1."),
+        ("out_layers.0.", "norm2."),
+        ("out_layers.3.", "conv2."),
+        ("emb_layers.1.", "time_emb_proj."),
+        ("skip_connection.", "conv_shortcut."),
+    ]
+
+    unet_conversion_map = []
+    for sd, hf in unet_conversion_map_layer:
+        if "resnets" in hf:
+            for sd_res, hf_res in unet_conversion_map_resnet:
+                unet_conversion_map.append((sd + sd_res, hf + hf_res))
+        else:
+            unet_conversion_map.append((sd, hf))
+
+    for j in range(2):
+        hf_time_embed_prefix = f"time_embedding.linear_{j+1}."
+        sd_time_embed_prefix = f"time_embed.{j*2}."
+        unet_conversion_map.append((sd_time_embed_prefix, hf_time_embed_prefix))
+
+    for j in range(2):
+        hf_label_embed_prefix = f"add_embedding.linear_{j+1}."
+        sd_label_embed_prefix = f"label_emb.0.{j*2}."
+        unet_conversion_map.append((sd_label_embed_prefix, hf_label_embed_prefix))
+
+    unet_conversion_map.append(("input_blocks.0.0.", "conv_in."))
+    unet_conversion_map.append(("out.0.", "conv_norm_out."))
+    unet_conversion_map.append(("out.2.", "conv_out."))
+
+    return unet_conversion_map
+
+
+SDXL_UNET_COMPVIS_MAP = {
+    f"{sd}".rstrip(".").replace(".", "_"): f"{hf}".rstrip(".").replace(".", "_")
+    for sd, hf in make_sdxl_unet_conversion_map()
+}

invokeai/backend/model_management/models/stable_diffusion.py

@@ -4,6 +4,7 @@ from enum import Enum
 from pydantic import Field
 from pathlib import Path
 from typing import Literal, Optional, Union
+from diffusers import StableDiffusionInpaintPipeline, StableDiffusionPipeline
 from .base import (
     ModelConfigBase,
     BaseModelType,
@@ -123,6 +124,7 @@ class StableDiffusion1Model(DiffusersModel):
             return _convert_ckpt_and_cache(
                 version=BaseModelType.StableDiffusion1,
                 model_config=config,
+                load_safety_checker=False,
                 output_path=output_path,
             )
         else:
@@ -259,9 +261,11 @@ def _convert_ckpt_and_cache(
     """
     app_config = InvokeAIAppConfig.get_config()
 
-    weights = app_config.root_path / model_config.path
+    weights = app_config.models_path / model_config.path
     config_file = app_config.root_path / model_config.config
     output_path = Path(output_path)
+    variant = model_config.variant
+    pipeline_class = StableDiffusionInpaintPipeline if variant == "inpaint" else StableDiffusionPipeline
 
     # return cached version if it exists
     if output_path.exists():
@@ -288,6 +292,7 @@ def _convert_ckpt_and_cache(
             original_config_file=config_file,
             extract_ema=True,
             scan_needed=True,
+            pipeline_class=pipeline_class,
             from_safetensors=weights.suffix == ".safetensors",
             precision=torch_dtype(choose_torch_device()),
             **kwargs,

invokeai/backend/model_management/models/stable_diffusion_onnx.py

@@ -0,0 +1,157 @@
+import os
+import json
+from enum import Enum
+from pydantic import Field
+from pathlib import Path
+from typing import Literal, Optional, Union
+from .base import (
+    ModelBase,
+    ModelConfigBase,
+    BaseModelType,
+    ModelType,
+    SubModelType,
+    ModelVariantType,
+    DiffusersModel,
+    SchedulerPredictionType,
+    SilenceWarnings,
+    read_checkpoint_meta,
+    classproperty,
+    OnnxRuntimeModel,
+    IAIOnnxRuntimeModel,
+)
+from invokeai.app.services.config import InvokeAIAppConfig
+
+
+class StableDiffusionOnnxModelFormat(str, Enum):
+    Olive = "olive"
+    Onnx = "onnx"
+
+
+class ONNXStableDiffusion1Model(DiffusersModel):
+    class Config(ModelConfigBase):
+        model_format: Literal[StableDiffusionOnnxModelFormat.Onnx]
+        variant: ModelVariantType
+
+    def __init__(self, model_path: str, base_model: BaseModelType, model_type: ModelType):
+        assert base_model == BaseModelType.StableDiffusion1
+        assert model_type == ModelType.ONNX
+        super().__init__(
+            model_path=model_path,
+            base_model=BaseModelType.StableDiffusion1,
+            model_type=ModelType.ONNX,
+        )
+
+        for child_name, child_type in self.child_types.items():
+            if child_type is OnnxRuntimeModel:
+                self.child_types[child_name] = IAIOnnxRuntimeModel
+
+            # TODO: check that no optimum models provided
+
+    @classmethod
+    def probe_config(cls, path: str, **kwargs):
+        model_format = cls.detect_format(path)
+        in_channels = 4  # TODO:
+
+        if in_channels == 9:
+            variant = ModelVariantType.Inpaint
+        elif in_channels == 4:
+            variant = ModelVariantType.Normal
+        else:
+            raise Exception("Unkown stable diffusion 1.* model format")
+
+        return cls.create_config(
+            path=path,
+            model_format=model_format,
+            variant=variant,
+        )
+
+    @classproperty
+    def save_to_config(cls) -> bool:
+        return True
+
+    @classmethod
+    def detect_format(cls, model_path: str):
+        # TODO: Detect onnx vs olive
+        return StableDiffusionOnnxModelFormat.Onnx
+
+    @classmethod
+    def convert_if_required(
+        cls,
+        model_path: str,
+        output_path: str,
+        config: ModelConfigBase,
+        base_model: BaseModelType,
+    ) -> str:
+        return model_path
+
+
+class ONNXStableDiffusion2Model(DiffusersModel):
+    # TODO: check that configs overwriten properly
+    class Config(ModelConfigBase):
+        model_format: Literal[StableDiffusionOnnxModelFormat.Onnx]
+        variant: ModelVariantType
+        prediction_type: SchedulerPredictionType
+        upcast_attention: bool
+
+    def __init__(self, model_path: str, base_model: BaseModelType, model_type: ModelType):
+        assert base_model == BaseModelType.StableDiffusion2
+        assert model_type == ModelType.ONNX
+        super().__init__(
+            model_path=model_path,
+            base_model=BaseModelType.StableDiffusion2,
+            model_type=ModelType.ONNX,
+        )
+
+        for child_name, child_type in self.child_types.items():
+            if child_type is OnnxRuntimeModel:
+                self.child_types[child_name] = IAIOnnxRuntimeModel
+            # TODO: check that no optimum models provided
+
+    @classmethod
+    def probe_config(cls, path: str, **kwargs):
+        model_format = cls.detect_format(path)
+        in_channels = 4  # TODO:
+
+        if in_channels == 9:
+            variant = ModelVariantType.Inpaint
+        elif in_channels == 5:
+            variant = ModelVariantType.Depth
+        elif in_channels == 4:
+            variant = ModelVariantType.Normal
+        else:
+            raise Exception("Unkown stable diffusion 2.* model format")
+
+        if variant == ModelVariantType.Normal:
+            prediction_type = SchedulerPredictionType.VPrediction
+            upcast_attention = True
+
+        else:
+            prediction_type = SchedulerPredictionType.Epsilon
+            upcast_attention = False
+
+        return cls.create_config(
+            path=path,
+            model_format=model_format,
+            variant=variant,
+            prediction_type=prediction_type,
+            upcast_attention=upcast_attention,
+        )
+
+    @classproperty
+    def save_to_config(cls) -> bool:
+        return True
+
+    @classmethod
+    def detect_format(cls, model_path: str):
+        # TODO: Detect onnx vs olive
+        return StableDiffusionOnnxModelFormat.Onnx
+
+    @classmethod
+    def convert_if_required(
+        cls,
+        model_path: str,
+        output_path: str,
+        config: ModelConfigBase,
+        base_model: BaseModelType,
+    ) -> str:
+        return model_path

invokeai/backend/model_management/models/vae.py

@@ -1,9 +1,14 @@
 import os
-import torch
-import safetensors
 from enum import Enum
 from pathlib import Path
-from typing import Optional, Union, Literal
+from typing import Optional
+
+import safetensors
+import torch
+from diffusers.utils import is_safetensors_available
+from omegaconf import OmegaConf
+
+from invokeai.app.services.config import InvokeAIAppConfig
 from .base import (
     ModelBase,
     ModelConfigBase,
@@ -18,9 +23,6 @@ from .base import (
     InvalidModelException,
     ModelNotFoundException,
 )
-from invokeai.app.services.config import InvokeAIAppConfig
-from diffusers.utils import is_safetensors_available
-from omegaconf import OmegaConf
 
 
 class VaeModelFormat(str, Enum):
@@ -80,7 +82,7 @@ class VaeModel(ModelBase):
     @classmethod
     def detect_format(cls, path: str):
         if not os.path.exists(path):
-            raise ModelNotFoundException()
+            raise ModelNotFoundException(f"Does not exist as local file: {path}")
 
         if os.path.isdir(path):
             if os.path.exists(os.path.join(path, "config.json")):

invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py

@@ -78,10 +78,9 @@ class InvokeAIDiffuserComponent:
         self.cross_attention_control_context = None
         self.sequential_guidance = config.sequential_guidance
 
-    @classmethod
     @contextmanager
     def custom_attention_context(
-        cls,
+        self,
         unet: UNet2DConditionModel,  # note: also may futz with the text encoder depending on requested LoRAs
         extra_conditioning_info: Optional[ExtraConditioningInfo],
         step_count: int,
@@ -91,18 +90,19 @@ class InvokeAIDiffuserComponent:
             old_attn_processors = unet.attn_processors
             # Load lora conditions into the model
             if extra_conditioning_info.wants_cross_attention_control:
-                cross_attention_control_context = Context(
+                self.cross_attention_control_context = Context(
                     arguments=extra_conditioning_info.cross_attention_control_args,
                     step_count=step_count,
                 )
                 setup_cross_attention_control_attention_processors(
                     unet,
-                    cross_attention_control_context,
+                    self.cross_attention_control_context,
                 )
 
         try:
             yield None
         finally:
+            self.cross_attention_control_context = None
             if old_attn_processors is not None:
                 unet.set_attn_processor(old_attn_processors)
             # TODO resuscitate attention map saving

invokeai/backend/util/devices.py

@@ -1,6 +1,8 @@
 from __future__ import annotations
 
 from contextlib import nullcontext
+from packaging import version
+import platform
 
 import torch
 from torch import autocast
@@ -30,7 +32,7 @@ def choose_precision(device: torch.device) -> str:
         device_name = torch.cuda.get_device_name(device)
         if not ("GeForce GTX 1660" in device_name or "GeForce GTX 1650" in device_name):
             return "float16"
-    elif device.type == "mps":
+    elif device.type == "mps" and version.parse(platform.mac_ver()[0]) < version.parse("14.0.0"):
         return "float16"
     return "float32"
 

invokeai/frontend/install/invokeai_update.py

@@ -112,7 +112,7 @@ def main():
 
     extras = get_extras()
 
-    print(f":crossed_fingers: Upgrading to [yellow]{tag if tag else release}[/yellow]")
+    print(f":crossed_fingers: Upgrading to [yellow]{tag or release or branch}[/yellow]")
     if release:
         cmd = f'pip install "invokeai{extras} @ {INVOKE_AI_SRC}/{release}.zip" --use-pep517 --upgrade'
     elif tag:

invokeai/frontend/install/model_install.py

@@ -58,6 +58,9 @@ logger = InvokeAILogger.getLogger()
 # from https://stackoverflow.com/questions/92438/stripping-non-printable-characters-from-a-string-in-python
 NOPRINT_TRANS_TABLE = {i: None for i in range(0, sys.maxunicode + 1) if not chr(i).isprintable()}
 
+# maximum number of installed models we can display before overflowing vertically
+MAX_OTHER_MODELS = 72
+
 
 def make_printable(s: str) -> str:
     """Replace non-printable characters in a string"""
@@ -102,7 +105,7 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
             SingleSelectColumns,
             values=[
                 "STARTER MODELS",
-                "MORE MODELS",
+                "MAIN MODELS",
                 "CONTROLNETS",
                 "LORA/LYCORIS",
                 "TEXTUAL INVERSION",
@@ -153,7 +156,7 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
             BufferBox,
             name="Log Messages",
             editable=False,
-            max_height=8,
+            max_height=15,
         )
 
         self.nextrely += 1
@@ -253,6 +256,7 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
         model_labels = [self.model_labels[x] for x in model_list]
 
         show_recommended = len(self.installed_models) == 0
+        truncated = False
         if len(model_list) > 0:
             max_width = max([len(x) for x in model_labels])
             columns = window_width // (max_width + 8)  # 8 characters for "[x] " and padding
@@ -271,6 +275,10 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
                 )
             )
 
+            if len(model_labels) > MAX_OTHER_MODELS:
+                model_labels = model_labels[0:MAX_OTHER_MODELS]
+                truncated = True
+
             widgets.update(
                 models_selected=self.add_widget_intelligent(
                     MultiSelectColumns,
@@ -289,6 +297,16 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
                 models=model_list,
             )
 
+        if truncated:
+            widgets.update(
+                warning_message=self.add_widget_intelligent(
+                    npyscreen.FixedText,
+                    value=f"Too many models to display (max={MAX_OTHER_MODELS}). Some are not displayed.",
+                    editable=False,
+                    color="CAUTION",
+                )
+            )
+
         self.nextrely += 1
         widgets.update(
             download_ids=self.add_widget_intelligent(
@@ -313,7 +331,7 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
         widgets = self.add_model_widgets(
             model_type=model_type,
             window_width=window_width,
-            install_prompt=f"Additional {model_type.value.title()} models already installed.",
+            install_prompt=f"Installed {model_type.value.title()} models. Unchecked models in the InvokeAI root directory will be deleted. Enter URLs, paths or repo_ids to import.",
             **kwargs,
         )
 
@@ -399,7 +417,7 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
         self.ok_button.hidden = True
         self.display()
 
-        # for communication with the subprocess
+        # TO DO: Spawn a worker thread, not a subprocess
         parent_conn, child_conn = Pipe()
         p = Process(
             target=process_and_execute,
@@ -414,7 +432,6 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
         self.subprocess_connection = parent_conn
         self.subprocess = p
         app.install_selections = InstallSelections()
-        # process_and_execute(app.opt, app.install_selections)
 
     def on_back(self):
         self.parentApp.switchFormPrevious()
@@ -489,8 +506,6 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
 
         # rebuild the form, saving and restoring some of the fields that need to be preserved.
         saved_messages = self.monitor.entry_widget.values
-        # autoload_dir = str(config.root_path / self.pipeline_models['autoload_directory'].value)
-        # autoscan = self.pipeline_models['autoscan_on_startup'].value
 
         app.main_form = app.addForm(
             "MAIN",
@@ -544,12 +559,6 @@ class addModelsForm(CyclingForm, npyscreen.FormMultiPage):
             if downloads := section.get("download_ids"):
                 selections.install_models.extend(downloads.value.split())
 
-        # load directory and whether to scan on startup
-        # if self.parentApp.autoload_pending:
-        #     selections.scan_directory = str(config.root_path / self.pipeline_models['autoload_directory'].value)
-        #     self.parentApp.autoload_pending = False
-        # selections.autoscan_on_startup = self.pipeline_models['autoscan_on_startup'].value
-
 
 class AddModelApplication(npyscreen.NPSAppManaged):
     def __init__(self, opt):
@@ -639,6 +648,11 @@ def process_and_execute(
     selections: InstallSelections,
     conn_out: Connection = None,
 ):
+    # need to reinitialize config in subprocess
+    config = InvokeAIAppConfig.get_config()
+    args = ["--root", opt.root] if opt.root else []
+    config.parse_args(args)
+
     # set up so that stderr is sent to conn_out
     if conn_out:
         translator = StderrToMessage(conn_out)
@@ -656,38 +670,11 @@ def process_and_execute(
         conn_out.close()
 
 
-def do_listings(opt) -> bool:
-    """List installed models of various sorts, and return
-    True if any were requested."""
-    model_manager = ModelManager(config.model_conf_path)
-    if opt.list_models == "diffusers":
-        print("Diffuser models:")
-        model_manager.print_models()
-    elif opt.list_models == "controlnets":
-        print("Installed Controlnet Models:")
-        cnm = model_manager.list_controlnet_models()
-        print(textwrap.indent("\n".join([x for x in cnm if cnm[x]]), prefix="   "))
-    elif opt.list_models == "loras":
-        print("Installed LoRA/LyCORIS Models:")
-        cnm = model_manager.list_lora_models()
-        print(textwrap.indent("\n".join([x for x in cnm if cnm[x]]), prefix="   "))
-    elif opt.list_models == "tis":
-        print("Installed Textual Inversion Embeddings:")
-        cnm = model_manager.list_ti_models()
-        print(textwrap.indent("\n".join([x for x in cnm if cnm[x]]), prefix="   "))
-    else:
-        return False
-    return True
-
-
 # --------------------------------------------------------
 def select_and_download_models(opt: Namespace):
     precision = "float32" if opt.full_precision else choose_precision(torch.device(choose_torch_device()))
     config.precision = precision
     helper = lambda x: ask_user_for_prediction_type(x)
-    # if do_listings(opt):
-    # pass
-
     installer = ModelInstall(config, prediction_type_helper=helper)
     if opt.list_models:
         installer.list_models(opt.list_models)
@@ -706,8 +693,6 @@ def select_and_download_models(opt: Namespace):
         # needed to support the probe() method running under a subprocess
         torch.multiprocessing.set_start_method("spawn")
 
-        # the third argument is needed in the Windows 11 environment in
-        # order to launch and resize a console window running this program
         set_min_terminal_size(MIN_COLS, MIN_LINES)
         installApp = AddModelApplication(opt)
         try:

invokeai/frontend/merge/merge_diffusers.py

@@ -320,7 +320,7 @@ class mergeModelsForm(npyscreen.FormMultiPageAction):
 
     def get_model_names(self, base_model: BaseModelType = None) -> List[str]:
         model_names = [
-            info["name"]
+            info["model_name"]
             for info in self.model_manager.list_models(model_type=ModelType.Main, base_model=base_model)
             if info["model_format"] == "diffusers"
         ]

invokeai/frontend/web/config/vite.package.config.ts

@@ -21,7 +21,7 @@ export const packageConfig: UserConfig = {
       fileName: (format) => `invoke-ai-ui.${format}.js`,
     },
     rollupOptions: {
-      external: ['react', 'react-dom', '@emotion/react'],
+      external: ['react', 'react-dom', '@emotion/react', '@chakra-ui/react'],
       output: {
         globals: {
           react: 'React',

invokeai/frontend/web/dist/index.html

@@ -12,7 +12,7 @@
         margin: 0;
       }
     </style>
-    <script type="module" crossorigin src="./assets/index-bad7ff83.js"></script>
+    <script type="module" crossorigin src="./assets/index-de589048.js"></script>
   </head>
 
   <body dir="ltr">

invokeai/frontend/web/dist/locales/en.json

@@ -124,7 +124,8 @@
         "deleteImageBin": "Deleted images will be sent to your operating system's Bin.",
         "deleteImagePermanent": "Deleted images cannot be restored.",
         "images": "Images",
-        "assets": "Assets"
+        "assets": "Assets",
+        "autoAssignBoardOnClick": "Auto-Assign Board on Click"
     },
     "hotkeys": {
         "keyboardShortcuts": "Keyboard Shortcuts",
@@ -342,6 +343,8 @@
         "diffusersModels": "Diffusers",
         "loraModels": "LoRAs",
         "safetensorModels": "SafeTensors",
+        "onnxModels": "Onnx",
+        "oliveModels": "Olives",
         "modelAdded": "Model Added",
         "modelUpdated": "Model Updated",
         "modelUpdateFailed": "Model Update Failed",

invokeai/frontend/web/package.json

@@ -23,7 +23,7 @@
     "dev": "concurrently \"vite dev\" \"yarn run theme:watch\"",
     "dev:host": "concurrently \"vite dev --host\" \"yarn run theme:watch\"",
     "build": "yarn run lint && vite build",
-    "typegen": "npx ts-node scripts/typegen.ts",
+    "typegen": "node scripts/typegen.js",
     "preview": "vite preview",
     "lint:madge": "madge --circular src/main.tsx",
     "lint:eslint": "eslint --max-warnings=0 .",
@@ -116,6 +116,7 @@
   },
   "peerDependencies": {
     "@chakra-ui/cli": "^2.4.0",
+    "@chakra-ui/react": "^2.8.0",
     "react": "^18.2.0",
     "react-dom": "^18.2.0",
     "ts-toolbelt": "^9.6.0"

invokeai/frontend/web/public/locales/en.json

@@ -124,7 +124,8 @@
         "deleteImageBin": "Deleted images will be sent to your operating system's Bin.",
         "deleteImagePermanent": "Deleted images cannot be restored.",
         "images": "Images",
-        "assets": "Assets"
+        "assets": "Assets",
+        "autoAssignBoardOnClick": "Auto-Assign Board on Click"
     },
     "hotkeys": {
         "keyboardShortcuts": "Keyboard Shortcuts",
@@ -342,6 +343,8 @@
         "diffusersModels": "Diffusers",
         "loraModels": "LoRAs",
         "safetensorModels": "SafeTensors",
+        "onnxModels": "Onnx",
+        "oliveModels": "Olives",
         "modelAdded": "Model Added",
         "modelUpdated": "Model Updated",
         "modelUpdateFailed": "Model Update Failed",

invokeai/frontend/web/src/app/components/App.tsx

@@ -4,8 +4,9 @@ import { appStarted } from 'app/store/middleware/listenerMiddleware/listeners/ap
 import { useAppDispatch, useAppSelector } from 'app/store/storeHooks';
 import { PartialAppConfig } from 'app/types/invokeai';
 import ImageUploader from 'common/components/ImageUploader';
+import ChangeBoardModal from 'features/changeBoardModal/components/ChangeBoardModal';
+import DeleteImageModal from 'features/deleteImageModal/components/DeleteImageModal';
 import GalleryDrawer from 'features/gallery/components/GalleryPanel';
-import DeleteImageModal from 'features/imageDeletion/components/DeleteImageModal';
 import SiteHeader from 'features/system/components/SiteHeader';
 import { configChanged } from 'features/system/store/configSlice';
 import { languageSelector } from 'features/system/store/systemSelectors';
@@ -16,7 +17,6 @@ import ParametersDrawer from 'features/ui/components/ParametersDrawer';
 import i18n from 'i18n';
 import { size } from 'lodash-es';
 import { ReactNode, memo, useEffect } from 'react';
-import UpdateImageBoardModal from '../../features/gallery/components/Boards/UpdateImageBoardModal';
 import GlobalHotkeys from './GlobalHotkeys';
 import Toaster from './Toaster';
 
@@ -84,7 +84,7 @@ const App = ({ config = DEFAULT_CONFIG, headerComponent }: Props) => {
         </Portal>
       </Grid>
       <DeleteImageModal />
-      <UpdateImageBoardModal />
+      <ChangeBoardModal />
       <Toaster />
       <GlobalHotkeys />
     </>

invokeai/frontend/web/src/app/components/ImageDnd/DragPreview.tsx

@@ -58,7 +58,7 @@ const DragPreview = (props: OverlayDragImageProps) => {
     );
   }
 
-  if (props.dragData.payloadType === 'IMAGE_NAMES') {
+  if (props.dragData.payloadType === 'IMAGE_DTOS') {
     return (
       <Flex
         sx={{
@@ -71,7 +71,7 @@ const DragPreview = (props: OverlayDragImageProps) => {
           ...STYLES,
         }}
       >
-        <Heading>{props.dragData.payload.image_names.length}</Heading>
+        <Heading>{props.dragData.payload.imageDTOs.length}</Heading>
         <Heading size="sm">Images</Heading>
       </Flex>
     );

invokeai/frontend/web/src/app/components/ImageDnd/ImageDndContext.tsx

@@ -18,27 +18,32 @@ import {
   DragStartEvent,
   TypesafeDraggableData,
 } from './typesafeDnd';
+import { logger } from 'app/logging/logger';
 
 type ImageDndContextProps = PropsWithChildren;
 
 const ImageDndContext = (props: ImageDndContextProps) => {
   const [activeDragData, setActiveDragData] =
     useState<TypesafeDraggableData | null>(null);
+  const log = logger('images');
 
   const dispatch = useAppDispatch();
 
-  const handleDragStart = useCallback((event: DragStartEvent) => {
-    console.log('dragStart', event.active.data.current);
-    const activeData = event.active.data.current;
-    if (!activeData) {
-      return;
-    }
-    setActiveDragData(activeData);
-  }, []);
+  const handleDragStart = useCallback(
+    (event: DragStartEvent) => {
+      log.trace({ dragData: event.active.data.current }, 'Drag started');
+      const activeData = event.active.data.current;
+      if (!activeData) {
+        return;
+      }
+      setActiveDragData(activeData);
+    },
+    [log]
+  );
 
   const handleDragEnd = useCallback(
     (event: DragEndEvent) => {
-      console.log('dragEnd', event.active.data.current);
+      log.trace({ dragData: event.active.data.current }, 'Drag ended');
       const overData = event.over?.data.current;
       if (!activeDragData || !overData) {
         return;
@@ -46,7 +51,7 @@ const ImageDndContext = (props: ImageDndContextProps) => {
       dispatch(dndDropped({ overData, activeData: activeDragData }));
       setActiveDragData(null);
     },
-    [activeDragData, dispatch]
+    [activeDragData, dispatch, log]
   );
 
   const mouseSensor = useSensor(MouseSensor, {

invokeai/frontend/web/src/app/components/ImageDnd/typesafeDnd.tsx

@@ -11,7 +11,6 @@ import {
   useDraggable as useOriginalDraggable,
   useDroppable as useOriginalDroppable,
 } from '@dnd-kit/core';
-import { BoardId } from 'features/gallery/store/types';
 import { ImageDTO } from 'services/api/types';
 
 type BaseDropData = {
@@ -54,9 +53,13 @@ export type AddToBatchDropData = BaseDropData & {
   actionType: 'ADD_TO_BATCH';
 };
 
-export type MoveBoardDropData = BaseDropData & {
-  actionType: 'MOVE_BOARD';
-  context: { boardId: BoardId };
+export type AddToBoardDropData = BaseDropData & {
+  actionType: 'ADD_TO_BOARD';
+  context: { boardId: string };
+};
+
+export type RemoveFromBoardDropData = BaseDropData & {
+  actionType: 'REMOVE_FROM_BOARD';
 };
 
 export type TypesafeDroppableData =
@@ -67,7 +70,8 @@ export type TypesafeDroppableData =
   | NodesImageDropData
   | AddToBatchDropData
   | NodesMultiImageDropData
-  | MoveBoardDropData;
+  | AddToBoardDropData
+  | RemoveFromBoardDropData;
 
 type BaseDragData = {
   id: string;
@@ -78,14 +82,12 @@ export type ImageDraggableData = BaseDragData & {
   payload: { imageDTO: ImageDTO };
 };
 
-export type ImageNamesDraggableData = BaseDragData & {
-  payloadType: 'IMAGE_NAMES';
-  payload: { image_names: string[] };
+export type ImageDTOsDraggableData = BaseDragData & {
+  payloadType: 'IMAGE_DTOS';
+  payload: { imageDTOs: ImageDTO[] };
 };
 
-export type TypesafeDraggableData =
-  | ImageDraggableData
-  | ImageNamesDraggableData;
+export type TypesafeDraggableData = ImageDraggableData | ImageDTOsDraggableData;
 
 interface UseDroppableTypesafeArguments
   extends Omit<UseDroppableArguments, 'data'> {
@@ -156,14 +158,39 @@ export const isValidDrop = (
     case 'SET_NODES_IMAGE':
       return payloadType === 'IMAGE_DTO';
     case 'SET_MULTI_NODES_IMAGE':
-      return payloadType === 'IMAGE_DTO' || 'IMAGE_NAMES';
+      return payloadType === 'IMAGE_DTO' || 'IMAGE_DTOS';
     case 'ADD_TO_BATCH':
-      return payloadType === 'IMAGE_DTO' || 'IMAGE_NAMES';
-    case 'MOVE_BOARD': {
+      return payloadType === 'IMAGE_DTO' || 'IMAGE_DTOS';
+    case 'ADD_TO_BOARD': {
       // If the board is the same, don't allow the drop
 
       // Check the payload types
-      const isPayloadValid = payloadType === 'IMAGE_DTO' || 'IMAGE_NAMES';
+      const isPayloadValid = payloadType === 'IMAGE_DTO' || 'IMAGE_DTOS';
+      if (!isPayloadValid) {
+        return false;
+      }
+
+      // Check if the image's board is the board we are dragging onto
+      if (payloadType === 'IMAGE_DTO') {
+        const { imageDTO } = active.data.current.payload;
+        const currentBoard = imageDTO.board_id ?? 'none';
+        const destinationBoard = overData.context.boardId;
+
+        return currentBoard !== destinationBoard;
+      }
+
+      if (payloadType === 'IMAGE_DTOS') {
+        // TODO (multi-select)
+        return true;
+      }
+
+      return false;
+    }
+    case 'REMOVE_FROM_BOARD': {
+      // If the board is the same, don't allow the drop
+
+      // Check the payload types
+      const isPayloadValid = payloadType === 'IMAGE_DTO' || 'IMAGE_DTOS';
       if (!isPayloadValid) {
         return false;
       }
@@ -172,20 +199,16 @@ export const isValidDrop = (
       if (payloadType === 'IMAGE_DTO') {
         const { imageDTO } = active.data.current.payload;
         const currentBoard = imageDTO.board_id;
-        const destinationBoard = overData.context.boardId;
 
-        const isSameBoard = currentBoard === destinationBoard;
-        const isDestinationValid = !currentBoard ? destinationBoard : true;
-
-        return !isSameBoard && isDestinationValid;
+        return currentBoard !== 'none';
       }
 
-      if (payloadType === 'IMAGE_NAMES') {
+      if (payloadType === 'IMAGE_DTOS') {
         // TODO (multi-select)
-        return false;
+        return true;
       }
 
-      return true;
+      return false;
     }
     default:
       return false;

invokeai/frontend/web/src/app/components/InvokeAIUI.tsx

@@ -1,3 +1,6 @@
+import { Middleware } from '@reduxjs/toolkit';
+import { store } from 'app/store/store';
+import { PartialAppConfig } from 'app/types/invokeai';
 import React, {
   lazy,
   memo,
@@ -6,18 +9,12 @@ import React, {
   useEffect,
 } from 'react';
 import { Provider } from 'react-redux';
-import { store } from 'app/store/store';
-
-import Loading from '../../common/components/Loading/Loading';
 import { addMiddleware, resetMiddlewares } from 'redux-dynamic-middlewares';
-import { PartialAppConfig } from 'app/types/invokeai';
-
-import '../../i18n';
+import { $authToken, $baseUrl, $projectId } from 'services/api/client';
 import { socketMiddleware } from 'services/events/middleware';
-import { Middleware } from '@reduxjs/toolkit';
+import Loading from '../../common/components/Loading/Loading';
+import '../../i18n';
 import ImageDndContext from './ImageDnd/ImageDndContext';
-import { AddImageToBoardContextProvider } from '../contexts/AddImageToBoardContext';
-import { $authToken, $baseUrl } from 'services/api/client';
 
 const App = lazy(() => import('./App'));
 const ThemeLocaleProvider = lazy(() => import('./ThemeLocaleProvider'));
@@ -28,6 +25,7 @@ interface Props extends PropsWithChildren {
   config?: PartialAppConfig;
   headerComponent?: ReactNode;
   middleware?: Middleware[];
+  projectId?: string;
 }
 
 const InvokeAIUI = ({
@@ -36,6 +34,7 @@ const InvokeAIUI = ({
   config,
   headerComponent,
   middleware,
+  projectId,
 }: Props) => {
   useEffect(() => {
     // configure API client token
@@ -48,6 +47,11 @@ const InvokeAIUI = ({
       $baseUrl.set(apiUrl);
     }
 
+    // configure API client project header
+    if (projectId) {
+      $projectId.set(projectId);
+    }
+
     // reset dynamically added middlewares
     resetMiddlewares();
 
@@ -67,8 +71,9 @@ const InvokeAIUI = ({
       // Reset the API client token and base url on unmount
       $baseUrl.set(undefined);
       $authToken.set(undefined);
+      $projectId.set(undefined);
     };
-  }, [apiUrl, token, middleware]);
+  }, [apiUrl, token, middleware, projectId]);
 
   return (
     <React.StrictMode>
@@ -76,9 +81,7 @@ const InvokeAIUI = ({
         <React.Suspense fallback={<Loading />}>
           <ThemeLocaleProvider>
             <ImageDndContext>
-              <AddImageToBoardContextProvider>
-                <App config={config} headerComponent={headerComponent} />
-              </AddImageToBoardContextProvider>
+              <App config={config} headerComponent={headerComponent} />
             </ImageDndContext>
           </ThemeLocaleProvider>
         </React.Suspense>

invokeai/frontend/web/src/app/contexts/AddImageToBoardContext.tsx

@@ -1,91 +0,0 @@
-import { useDisclosure } from '@chakra-ui/react';
-import { PropsWithChildren, createContext, useCallback, useState } from 'react';
-import { ImageDTO } from 'services/api/types';
-import { imagesApi } from 'services/api/endpoints/images';
-import { useAppDispatch } from '../store/storeHooks';
-
-export type ImageUsage = {
-  isInitialImage: boolean;
-  isCanvasImage: boolean;
-  isNodesImage: boolean;
-  isControlNetImage: boolean;
-};
-
-type AddImageToBoardContextValue = {
-  /**
-   * Whether the move image dialog is open.
-   */
-  isOpen: boolean;
-  /**
-   * Closes the move image dialog.
-   */
-  onClose: () => void;
-  /**
-   * The image pending movement
-   */
-  image?: ImageDTO;
-  onClickAddToBoard: (image: ImageDTO) => void;
-  handleAddToBoard: (boardId: string) => void;
-};
-
-export const AddImageToBoardContext =
-  createContext<AddImageToBoardContextValue>({
-    isOpen: false,
-    onClose: () => undefined,
-    onClickAddToBoard: () => undefined,
-    handleAddToBoard: () => undefined,
-  });
-
-type Props = PropsWithChildren;
-
-export const AddImageToBoardContextProvider = (props: Props) => {
-  const [imageToMove, setImageToMove] = useState<ImageDTO>();
-  const { isOpen, onOpen, onClose } = useDisclosure();
-  const dispatch = useAppDispatch();
-
-  // Clean up after deleting or dismissing the modal
-  const closeAndClearImageToDelete = useCallback(() => {
-    setImageToMove(undefined);
-    onClose();
-  }, [onClose]);
-
-  const onClickAddToBoard = useCallback(
-    (image?: ImageDTO) => {
-      if (!image) {
-        return;
-      }
-      setImageToMove(image);
-      onOpen();
-    },
-    [setImageToMove, onOpen]
-  );
-
-  const handleAddToBoard = useCallback(
-    (boardId: string) => {
-      if (imageToMove) {
-        dispatch(
-          imagesApi.endpoints.addImageToBoard.initiate({
-            imageDTO: imageToMove,
-            board_id: boardId,
-          })
-        );
-        closeAndClearImageToDelete();
-      }
-    },
-    [dispatch, closeAndClearImageToDelete, imageToMove]
-  );
-
-  return (
-    <AddImageToBoardContext.Provider
-      value={{
-        isOpen,
-        image: imageToMove,
-        onClose: closeAndClearImageToDelete,
-        onClickAddToBoard,
-        handleAddToBoard,
-      }}
-    >
-      {props.children}
-    </AddImageToBoardContext.Provider>
-  );
-};

invokeai/frontend/web/src/app/contexts/ImageUploaderTriggerContext.ts

@@ -1,8 +0,0 @@
-import { createContext } from 'react';
-
-type VoidFunc = () => void;
-
-type ImageUploaderTriggerContextType = VoidFunc | null;
-
-export const ImageUploaderTriggerContext =
-  createContext<ImageUploaderTriggerContextType>(null);

invokeai/frontend/web/src/app/store/enhancers/reduxRemember/serialize.ts

@@ -23,6 +23,6 @@ const serializationDenylist: {
 };
 
 export const serialize: SerializeFunction = (data, key) => {
-  const result = omit(data, serializationDenylist[key]);
+  const result = omit(data, serializationDenylist[key] ?? []);
   return JSON.stringify(result);
 };

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/index.ts

@@ -27,7 +27,8 @@ import {
   addImageDeletedFulfilledListener,
   addImageDeletedPendingListener,
   addImageDeletedRejectedListener,
-  addRequestedImageDeletionListener,
+  addRequestedSingleImageDeletionListener,
+  addRequestedMultipleImageDeletionListener,
 } from './listeners/imageDeleted';
 import { addImageDroppedListener } from './listeners/imageDropped';
 import {
@@ -111,7 +112,8 @@ addImageUploadedRejectedListener();
 addInitialImageSelectedListener();
 
 // Image deleted
-addRequestedImageDeletionListener();
+addRequestedSingleImageDeletionListener();
+addRequestedMultipleImageDeletionListener();
 addImageDeletedPendingListener();
 addImageDeletedFulfilledListener();
 addImageDeletedRejectedListener();

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/addFirstListImagesListener.ts.ts

@@ -1,12 +1,10 @@
 import { createAction } from '@reduxjs/toolkit';
 import { imageSelected } from 'features/gallery/store/gallerySlice';
 import { IMAGE_CATEGORIES } from 'features/gallery/store/types';
-import {
-  ImageCache,
-  getListImagesUrl,
-  imagesApi,
-} from 'services/api/endpoints/images';
+import { imagesApi } from 'services/api/endpoints/images';
 import { startAppListening } from '..';
+import { getListImagesUrl, imagesAdapter } from 'services/api/util';
+import { ImageCache } from 'services/api/types';
 
 export const appStarted = createAction('app/appStarted');
 
@@ -34,7 +32,8 @@ export const addFirstListImagesListener = () => {
 
       if (data.ids.length > 0) {
         // Select the first image
-        dispatch(imageSelected(data.ids[0] as string));
+        const firstImage = imagesAdapter.getSelectors().selectAll(data)[0];
+        dispatch(imageSelected(firstImage ?? null));
       }
     },
   });

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/appConfigReceived.ts

@@ -18,7 +18,9 @@ export const addAppConfigReceivedListener = () => {
       const infillMethod = getState().generation.infillMethod;
 
       if (!infill_methods.includes(infillMethod)) {
-        dispatch(setInfillMethod(infill_methods[0]));
+        // if there is no infill method, set it to the first one
+        // if there is no first one... god help us
+        dispatch(setInfillMethod(infill_methods[0] as string));
       }
 
       if (!nsfw_methods.includes('nsfw_checker')) {

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/boardAndImagesDeleted.ts

@@ -1,14 +1,14 @@
 import { resetCanvas } from 'features/canvas/store/canvasSlice';
 import { controlNetReset } from 'features/controlNet/store/controlNetSlice';
-import { getImageUsage } from 'features/imageDeletion/store/imageDeletionSelectors';
+import { getImageUsage } from 'features/deleteImageModal/store/selectors';
 import { nodeEditorReset } from 'features/nodes/store/nodesSlice';
 import { clearInitialImage } from 'features/parameters/store/generationSlice';
+import { imagesApi } from 'services/api/endpoints/images';
 import { startAppListening } from '..';
-import { boardsApi } from '../../../../../services/api/endpoints/boards';
 
 export const addDeleteBoardAndImagesFulfilledListener = () => {
   startAppListening({
-    matcher: boardsApi.endpoints.deleteBoardAndImages.matchFulfilled,
+    matcher: imagesApi.endpoints.deleteBoardAndImages.matchFulfilled,
     effect: async (action, { dispatch, getState }) => {
       const { deleted_images } = action.payload;
 

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/boardIdSelected.ts

@@ -10,6 +10,7 @@ import {
 } from 'features/gallery/store/types';
 import { imagesApi } from 'services/api/endpoints/images';
 import { startAppListening } from '..';
+import { imagesSelectors } from 'services/api/util';
 
 export const addBoardIdSelectedListener = () => {
   startAppListening({
@@ -52,8 +53,9 @@ export const addBoardIdSelectedListener = () => {
           queryArgs
         )(getState());
 
-        if (boardImagesData?.ids.length) {
-          dispatch(imageSelected((boardImagesData.ids[0] as string) ?? null));
+        if (boardImagesData) {
+          const firstImage = imagesSelectors.selectAll(boardImagesData)[0];
+          dispatch(imageSelected(firstImage ?? null));
         } else {
           // board has no images - deselect
           dispatch(imageSelected(null));

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/canvasSavedToGallery.ts

@@ -26,6 +26,8 @@ export const addCanvasSavedToGalleryListener = () => {
         return;
       }
 
+      const { autoAddBoardId } = state.gallery;
+
       dispatch(
         imagesApi.endpoints.uploadImage.initiate({
           file: new File([blob], 'savedCanvas.png', {
@@ -33,7 +35,7 @@ export const addCanvasSavedToGalleryListener = () => {
           }),
           image_category: 'general',
           is_intermediate: false,
-          board_id: state.gallery.autoAddBoardId,
+          board_id: autoAddBoardId === 'none' ? undefined : autoAddBoardId,
           crop_visible: true,
           postUploadAction: {
             type: 'TOAST',

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/controlNetAutoProcess.ts

@@ -31,15 +31,20 @@ const predicate: AnyListenerPredicate<RootState> = (
     // do not process if the user just disabled auto-config
     if (
       prevState.controlNet.controlNets[action.payload.controlNetId]
-        .shouldAutoConfig === true
+        ?.shouldAutoConfig === true
     ) {
       return false;
     }
   }
 
-  const { controlImage, processorType, shouldAutoConfig } =
-    state.controlNet.controlNets[action.payload.controlNetId];
+  const cn = state.controlNet.controlNets[action.payload.controlNetId];
 
+  if (!cn) {
+    // something is wrong, the controlNet should exist
+    return false;
+  }
+
+  const { controlImage, processorType, shouldAutoConfig } = cn;
   if (controlNetModelChanged.match(action) && !shouldAutoConfig) {
     // do not process if the action is a model change but the processor settings are dirty
     return false;

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/controlNetImageProcessed.ts

@@ -17,7 +17,7 @@ export const addControlNetImageProcessedListener = () => {
       const { controlNetId } = action.payload;
       const controlNet = getState().controlNet.controlNets[controlNetId];
 
-      if (!controlNet.controlImage) {
+      if (!controlNet?.controlImage) {
         log.error('Unable to process ControlNet image');
         return;
       }

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/imageDeleted.ts

@@ -1,57 +1,72 @@
 import { logger } from 'app/logging/logger';
 import { resetCanvas } from 'features/canvas/store/canvasSlice';
 import { controlNetReset } from 'features/controlNet/store/controlNetSlice';
+import { imageDeletionConfirmed } from 'features/deleteImageModal/store/actions';
+import { isModalOpenChanged } from 'features/deleteImageModal/store/slice';
 import { selectListImagesBaseQueryArgs } from 'features/gallery/store/gallerySelectors';
 import { imageSelected } from 'features/gallery/store/gallerySlice';
-import { imageDeletionConfirmed } from 'features/imageDeletion/store/actions';
-import { isModalOpenChanged } from 'features/imageDeletion/store/imageDeletionSlice';
 import { nodeEditorReset } from 'features/nodes/store/nodesSlice';
 import { clearInitialImage } from 'features/parameters/store/generationSlice';
 import { clamp } from 'lodash-es';
 import { api } from 'services/api';
 import { imagesApi } from 'services/api/endpoints/images';
+import { imagesAdapter } from 'services/api/util';
 import { startAppListening } from '..';
 
-/**
- * Called when the user requests an image deletion
- */
-export const addRequestedImageDeletionListener = () => {
+export const addRequestedSingleImageDeletionListener = () => {
   startAppListening({
     actionCreator: imageDeletionConfirmed,
     effect: async (action, { dispatch, getState, condition }) => {
-      const { imageDTO, imageUsage } = action.payload;
+      const { imageDTOs, imagesUsage } = action.payload;
+
+      if (imageDTOs.length !== 1 || imagesUsage.length !== 1) {
+        // handle multiples in separate listener
+        return;
+      }
+
+      const imageDTO = imageDTOs[0];
+      const imageUsage = imagesUsage[0];
+
+      if (!imageDTO || !imageUsage) {
+        // satisfy noUncheckedIndexedAccess
+        return;
+      }
 
       dispatch(isModalOpenChanged(false));
 
-      const { image_name } = imageDTO;
-
       const state = getState();
       const lastSelectedImage =
-        state.gallery.selection[state.gallery.selection.length - 1];
+        state.gallery.selection[state.gallery.selection.length - 1]?.image_name;
+
+      if (imageDTO && imageDTO?.image_name === lastSelectedImage) {
+        const { image_name } = imageDTO;
 
-      if (lastSelectedImage === image_name) {
         const baseQueryArgs = selectListImagesBaseQueryArgs(state);
         const { data } =
           imagesApi.endpoints.listImages.select(baseQueryArgs)(state);
 
-        const ids = data?.ids ?? [];
+        const cachedImageDTOs = data
+          ? imagesAdapter.getSelectors().selectAll(data)
+          : [];
 
-        const deletedImageIndex = ids.findIndex(
-          (result) => result.toString() === image_name
+        const deletedImageIndex = cachedImageDTOs.findIndex(
+          (i) => i.image_name === image_name
         );
 
-        const filteredIds = ids.filter((id) => id.toString() !== image_name);
+        const filteredImageDTOs = cachedImageDTOs.filter(
+          (i) => i.image_name !== image_name
+        );
 
         const newSelectedImageIndex = clamp(
           deletedImageIndex,
           0,
-          filteredIds.length - 1
+          filteredImageDTOs.length - 1
         );
 
-        const newSelectedImageId = filteredIds[newSelectedImageIndex];
+        const newSelectedImageDTO = filteredImageDTOs[newSelectedImageIndex];
 
-        if (newSelectedImageId) {
-          dispatch(imageSelected(newSelectedImageId as string));
+        if (newSelectedImageDTO) {
+          dispatch(imageSelected(newSelectedImageDTO));
         } else {
           dispatch(imageSelected(null));
         }
@@ -97,6 +112,66 @@ export const addRequestedImageDeletionListener = () => {
   });
 };
 
+/**
+ * Called when the user requests an image deletion
+ */
+export const addRequestedMultipleImageDeletionListener = () => {
+  startAppListening({
+    actionCreator: imageDeletionConfirmed,
+    effect: async (action, { dispatch, getState }) => {
+      const { imageDTOs, imagesUsage } = action.payload;
+
+      if (imageDTOs.length < 1 || imagesUsage.length < 1) {
+        // handle singles in separate listener
+        return;
+      }
+
+      try {
+        // Delete from server
+        await dispatch(
+          imagesApi.endpoints.deleteImages.initiate({ imageDTOs })
+        ).unwrap();
+        const state = getState();
+        const baseQueryArgs = selectListImagesBaseQueryArgs(state);
+        const { data } =
+          imagesApi.endpoints.listImages.select(baseQueryArgs)(state);
+
+        const newSelectedImageDTO = data
+          ? imagesAdapter.getSelectors().selectAll(data)[0]
+          : undefined;
+
+        if (newSelectedImageDTO) {
+          dispatch(imageSelected(newSelectedImageDTO));
+        } else {
+          dispatch(imageSelected(null));
+        }
+
+        dispatch(isModalOpenChanged(false));
+
+        // We need to reset the features where the image is in use - none of these work if their image(s) don't exist
+
+        if (imagesUsage.some((i) => i.isCanvasImage)) {
+          dispatch(resetCanvas());
+        }
+
+        if (imagesUsage.some((i) => i.isControlNetImage)) {
+          dispatch(controlNetReset());
+        }
+
+        if (imagesUsage.some((i) => i.isInitialImage)) {
+          dispatch(clearInitialImage());
+        }
+
+        if (imagesUsage.some((i) => i.isNodesImage)) {
+          dispatch(nodeEditorReset());
+        }
+      } catch {
+        // no-op
+      }
+    },
+  });
+};
+
 /**
  * Called when the actual delete request is sent to the server
  */