This is a WIP to add SDXL support.
Tasks:
- [x] SDXL model loading support
- [x] SDXL model installation
- [x] SDXL model loader
- [x] SDXL base invocations for text2latent and latent2latent
- [ ] SDXL refiner invocations for text2latent and latent2latent
- [x] Compel support / pooled embeddings
- [ ] Linear UI graph for SDXL
- [ ] Documentation
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [ ] Bug Fix
- [ ] Optimization
- [ ] Documentation Update
## Have you discussed this change with the InvokeAI team?
- [ ] Yes
- [ ] No, because:
## Description
## Related Tickets & Documents
<!--
For pull requests that relate or close an issue, please include them
below.
For example having the text: "closes #1234" would connect the current
pull
request to issue 1234. And when we merge the pull request, Github will
automatically close the issue.
-->
- Related Issue #
- Closes #
## QA Instructions, Screenshots, Recordings
<!--
Please provide steps on how to test changes, any hardware or
software specifications as well as any other pertinent information.
-->
## Added/updated tests?
- [ ] Yes
- [ ] No : _please replace this line with details on why tests
have not been included_
## [optional] Are there any post deployment tasks we need to perform?
fix json formatting to not have big red comment blocks
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [ ] Bug Fix
- [ ] Optimization
- [X] Documentation Update
## Have you discussed this change with the InvokeAI team?
- [ ] Yes
- [X] No, because: simple docs fix
## Description
Fix LOCAL_DEVELOPMENT.md json comment highlighting
## Related Tickets & Documents
<!--
For pull requests that relate or close an issue, please include them
below.
For example having the text: "closes #1234" would connect the current
pull
request to issue 1234. And when we merge the pull request, Github will
automatically close the issue.
-->
- Related Issue # n/a
- Closes # n/a
## QA Instructions, Screenshots, Recordings
<!--
Please provide steps on how to test changes, any hardware or
software specifications as well as any other pertinent information.
-->
## Added/updated tests?
- [ ] Yes
- [x] No : simple docs change
This PR completely ports over the Model Manager to 3.0 -- all of the
functionality has now been restored in addition to the following
changes.
- Model Manager now has been moved to its own tab on the left hand side.
- Model Manager has three tabs - Model Manager, Import Models and Merge
Models
- The edit forms for the Models now allow the users to update the model
name and the base model too along with other details.
- Checkpoint Edit form now displays the available config files from
InvokeAI and also allows users to supply their own custom config file.
- Under Import Models you can directly add models or a scan a folder for
your checkpoint files.
- Adding models has two modes -- Simple and Advanced.
- In Simple Mode, you just simply need to pass a path and InvokeAI will
try to determine kind of model it is and fill up the rest of the details
accordingly. This input lets you supply local paths to diffusers / local
paths to checkpoints / huggingface repo ID's to download models /
CivitAI links.
- Simple Mode also allows you to download different models types like
VAE's and Controlnet models and etc. Not just main models.
- In cases where the auto detection system of InvokeAI fails to read a
model correctly, you can take the manual approach and go to Advanced
where you can configure your model while adding it exactly the way you
want it. Both Diffusers and Checkpoint models now have their own custom
forms.
- Scan Models has been cleaned up. It will now only display the models
that are not already installed to InvokeAI. And each item will have two
options - Quick Add and Advanced .. replicating the Add Model behavior
from above.
- Scan Models now has a search bar for you to search through your
scanned models.
- Merge Models functionality has been restored.
This is a wrap for this PR.
**TODO: (Probably for 3.1)**
- Add model management for model types such as VAE's and ControlNet
Models
- Replace the VAE slot on the edit forms with the installed VAE drop
down + custom option
[feat(nodes): emit model loading
events](7b6159f8d6)
- remove dependency on having access to a `node` during emits, would
need a bit of additional args passed through the system and I don't
think its necessary at this point. this also allowed us to drop an
extraneous fetching/parsing of the session from db.
- provide the invocation context to all `get_model()` calls, so the
events are able to be emitted
- test all model loading events in the app and confirm socket events are
received
[feat(ui): add listeners for model load
events](c487166d9c)
- currently only exposed as DEBUG-level logs
---
One change I missed in the commit messages is the `ModelInfo` class is
not serializable, so I split out the pieces of information we didn't
already have (hash, location, precision) and added them to the event
payload directly.
This small patch improves the stability of `invokeai-*` scripts by
avoiding crashes in the model manager while scanning the models
directory for new and removed models.
