Bria is incompatible with latest Diffusers
BriaTransformer2DModel fails when using latest diffusers from main branch due to recent changes in FluxSingleTransformerBlock
and if bria cannot work with latest diffusers and requires a version-locked installs, its pretty much dead-on-arrival.
│/home/vlado/dev/sdnext/pipelines/bria/bria_pipeline.py:452 in __call__│
││
│❱ 452 │ │ │ │ noise_pred = self.transformer(│
│ 453 │ │ │ │ │ hidden_states=latent_model_input,│
...
TypeError: FluxSingleTransformerBlock.forward() missing 1 required positional argument: 'encoder_hidden_states'
one easy option is dont inherit from FluxSingleTransformerBlock, but make a copy of it before the breaking changes occured.
Hey
@vladmandic
, I looked into it and found that there are some breaking changes in the latest diffusers version.
I’m working on making us native in diffusers🤗 and I hope that in the next couple of days they will merge the PR and it will be way easier to use (and also compatible with old diffusers versions).
hi
@vladmandic
you can install diffusers from main and use our pipeline
for more complete example: https://github.com/Bria-AI/BRIA-3.2