- Attempting to run a prompt with a LoRA based on SD v1.X against a
model based on v2.X will now throw an
`IncompatibleModelException`. To import this exception:
`from ldm.modules.lora_manager import IncompatibleModelException`
(maybe this should be defined in ModelManager?)
- Enhance `LoraManager.list_loras()` to accept an optional integer
argument, `token_vector_length`. This will filter the returned LoRA
models to return only those that match the indicated length. Use:
```
768 => for models based on SD v1.X
1024 => for models based on SD v2.X
```
Note that this filtering requires each LoRA file to be opened
by `torch.safetensors`. It will take ~8s to scan a directory of
40 files.
- Added new static methods to `ldm.modules.kohya_lora_manager`:
- check_model_compatibility()
- vector_length_from_checkpoint()
- vector_length_from_checkpoint_file()
- Previously the user's preferred precision was used to select which
version branch of a diffusers model would be downloaded. Half-precision
would try to download the 'fp16' branch if it existed.
- Turns out that with waifu-diffusion this logic doesn't work, as
'fp16' gets you waifu-diffusion v1.3, while 'main' gets you
waifu-diffusion v1.4. Who knew?
- This PR adds a new optional "revision" field to `models.yaml`. This
can be used to override the diffusers branch version. In the case of
Waifu diffusion, INITIAL_MODELS.yaml now specifies the "main" branch.
- This PR also quenches the NSFW nag that downloading diffusers sometimes
triggers.
- Closes#3160
- Remove unused (and probably dangerous) `unload_applied_loras()` method
- Remove unused `LoraManager.loras_to_load` attribute
- Change default LoRA weight to 0.75 when using WebUI to add a LoRA to prompt.
- removed app directory (a 3.0 feature), so app tests had to go too
- fixed regular expression in the concepts lib which was causing deprecation warnings
Instructions:
1. Download LoRA .safetensors files of your choice and place in
`INVOKEAIROOT/loras`. Unlike the draft version of this, the file
names can contain underscores and alphanumerics. Names with
arbitrary unicode characters are not supported.
2. Add `withLora(lora-file-basename,weight)` to your prompt. The
weight is optional and will default to 1.0. A few examples, assuming
that a LoRA file named `loras/sushi.safetensors` is present:
```
family sitting at dinner table eating sushi withLora(sushi,0.9)
family sitting at dinner table eating sushi withLora(sushi, 0.75)
family sitting at dinner table eating sushi withLora(sushi)
```
Multiple `withLora()` prompt fragments are allowed. The weight can be
arbitrarily large, but the useful range is roughly 0.5 to 1.0. Higher
weights make the LoRA's influence stronger.
In my limited testing, I found it useful to reduce the CFG to avoid
oversharpening. Also I got better results when running the LoRA on top
of the model on which it was based during training.
Don't try to load a SD 1.x-trained LoRA into a SD 2.x model, and vice
versa. You will get a nasty stack trace. This needs to be cleaned up.
3. You can change the location of the `loras` directory by passing the
`--lora_directory` option to `invokeai.
Documentation can be found in docs/features/LORAS.md.
