Open shumoLR opened 1 month ago
what preprocessor are you using for inpainting? I doesn't seem to work for me?
I use the inaint processor, what processor should I use?
I tried using that and didn't have good result, just black result.
I think the problem is that in his example under the mask there are non-zero values, and if it was trained the same (not masking the input image, and only masking the control-img), then the model would fail during inference. But if you infill with 127 instead of 0, it starts to at least somehow work. Changing strength == 1 and using "combine-img" instead of "combine-noise" (aka mixing the clean img predictions, not the noised versions of them) results in almost perfect outputs
I have updated the inference scripts to fix the bug, please try it again, sorry to bring you inconvience.
In the original scripts, the promax just can't load normally.
Thank you very much for your selflessness. Now I have a problem. In the process of using [Image Outpainting], I found that the model used is ProMax, but [Image Inpainting] is perfectly usable, is it me who set the problem? Thank you for your reply.![image](https://github.com/user-attachments/assets/d6a366ab-da22-4000-8adc-e1872d443c0a)