Clip Skip breaks when you supply a number greater than the number of
layers for the model type. So capping this out based on the model on the
frontend
- `sd-1` at 12
- `sd-2` at 24
- Will update later to whatever SDXL needs if it is different.
- Also fixes LoRA's breaking with Clip Skip.
My PR to fix an issue with the handling of formdata in `openapi-fetch` is released. This means we no longer need to patch the package (no patches at all now!).
This PR bumps its version and adds a transformer to our typegen script to handle typing binary form fields correctly as `Blob`.
Also regens types.
This is PR adds the following API methods for managing models:
* list_models (GET)
* update_model (PATCH)
* import_model (POST)
* delete_model (DELETE)
* convert_model (PUT)
* merge_models (PUT)
* load images on gallery render
* wait for models to be loaded before you can invoke
---------
Co-authored-by: Mary Hipp <maryhipp@Marys-MacBook-Air.local>
This PR enables model manager importation of diffusers-style .bin LoRAs.
However, since there is no backend support for this type of LoRA yet,
attempts to use them will result in an unimplemented error.
It closes#3636 and #3637
The list models route should just be the base route path, and should use query parameters as opposed to path parameters (which cannot be optional)
Removed defaults for update model route - for the purposes of the API, we should always be explicit with this
This PR fixes the migrate script so that it uses the same directory for
both the tokenizer and text encoder CLIP models. This will fix a crash
that occurred during checkpoint->diffusers conversions
This PR also removes the check for an existing models directory in the
target root directory when `invokeai-migrate3` is run.
