Open crapthings opened 8 months ago
+1
Hi sorry for the delay, Thanks for your interest in diff diff. Actually, I now experimented with a modified version of the inpaint pipeline of diffusers and it seems to work. On each step, inpaint models take 9 channel (including masked image and binary mask), which are changed according to the current threshold. It is worth noting that the inpaint models tend to make more conservative changes. I need to think what is the support plan of this repo.
is this working with1.5 inpainting or sdxl inpainting?