Pipeline doesn't work with Diffusers=0.25.1 about auffusion

[happylittlecat2333]

They have similar performance since I trained them with same datasets.

[happylittlecat2333]

Thanks for your attention! Because my auffusion_pipeline try to explore different combination of text encoder, therefore I try to hack the default StableDiffusionPipeline and add these variables. On the other hand, I trained another model and it can use the default StableDiffusionPipeline in notebook, you can try to use this model in Diffusers==0.25.1 and it might work.

PS: the attention-based method in notebook depend on Diffusers=0.18, therefore method except attention-based might work fine in your case.

[IFICL]

Thanks for your reply. This is helpful. I wonder what the performance gap between Auffusion-Full-no-adapter and Auffusion-Full.