The 1.5 inpainting model is great! Significant improvement over how to do inpainting before. Highly recommended. Integrated into my mobile app last night and enjoying it so far!
any idea if you can merge a dreambooth checkpoint with the inpainting model to get the best of both? I mostly use inpainting to fix custom trained faces.
Yeah, that would be tricky to merge existing ones together. But you should be able to fine-tune the inpainting model the same way as you tune with dreambooth?
For faces, I haven't looked deep, but seems CodeFormer's training cost is minimal, should be able to fine-tune that model, probably better?