Add SD3 pipeline #438
Add SD3 pipeline #438
Conversation
|
This seems great, I tried it with both models and had no problems. All the "real work" is in apple/ml-stable-diffusion#329 and my inclination is to wait until that gets approved and merged before I merge this, but there's no indicator of how long that will take, so I might just go ahead with it, especially if I start to see the SD community start to release interesting SD3 models |
It was already merged 👍 |
Description:
Hi there - here is a PR to enable the new SD3 pipeline for MochiDiffusion. The models can be downloaded here:
512x512 no T5: https://huggingface.co/argmaxinc/coreml-stable-diffusion-3-medium/tree/main
1024x1024 with T5: https://huggingface.co/argmaxinc/coreml-stable-diffusion-3-medium-1024-t5/tree/main
These were generated with DiffusionKit in conjunction with ml-stable-diffusion for the shared text encoders.
Please note the the SD3 PR into ml-stable-diffusion is currently under review, and will need to be updated to main when merged apple/ml-stable-diffusion#329
Additionally, the image to image is not yet supported, so the app will just start from the beginning when using a starting image.