Open skyler14 opened 3 months ago
inpainting is totally possible, the workflow is the same as standard inpainting. I'll post a workflow when I have time
I get this error with in-painting models I made with comfy ((inpaint_model - base_model) * 1.0 + other_model):
RuntimeError: shape '[320, 9, 3, 3]' is invalid for input of size 11520
Edit: 11520 seems to be the tensor shape of [320,4,3,3], is it even possible to fit it with a differently shaped Tensor? it seem to be 5 additional input channels, 4 for the encoded masked-image and 1 for the mask itself. as per the sdxl model card
Edit2: NVM! I used the depth example and it seems to be related to 256 depth controlnet instead of instantid
inpainting is totally possible, the workflow is the same as standard inpainting. I'll post a workflow when I have time
any chance you might have some time
There was a custom node library preceding the release this one called comfyui instantid faceswap, which seemed to do a type of inpainting based around InstantID using any model (not limited to inpainting). Is it possible to recreate this with an workflow in this UI or are there missing modules that his use of the diffusers library allowed. Have you experimented with any inpainting pipelines and gooten similar results and if so can you add those to examples or attach here?