This model doesn't work correctly with the default pipeline
#3
by frutiemax - opened
I've finetuned a model based on this checkpoint and I've noticed the image get really noisy when introducing some elements that were not part of my dataset (and were part of the base checkpoint). I then tried to generate images with this checkpoint to see it doesn't work with the default pipeline i.e. the image is screwed up.
I then tried Efficient-Large-Model/Sana_1600M_1024px_MultiLing_diffusers, and this one works perfectly fine! Either update this model to work with the default pipeline or remove it.
Thanks man. The problem is caused by diffusers code. I pull a PR to fix it.
https://github.com/huggingface/diffusers/pull/10431