ComfyUI-TiledDiffusion icon indicating copy to clipboard operation
ComfyUI-TiledDiffusion copied to clipboard

how to do text-to-image chain tiled diffusion with depth and lineart? Workflow help

Open mikebilly opened this issue 6 months ago • 0 comments

Normally, I will take an input picture, chaining apply advanced controlnet which uses depth and apply advanced controlnet which uses lineart, to generate a new image using prompt. So basicall,y I want to generate a new image based on prompt and depth + lineart controlnet. The reason why I have to use advanced controlnet is it has attention masking (mask_optional), which is compulsory for me.

This problem is that when my input picture is not in the optimal aspect ratio or resolution, (512x512), it doesn't generate good images all the time, and when my input picture is very wide, it generates very ugly and illogical images.

So how can I use TiledDiffusion to solve this? I want to take in the input image at its original resolution, process the controlnet depth and lineart using 512x512 tiles, (to make sure that it's doing the best it can at originally trained resolution), also making sure that I can use attention masking (mask_optional), to generate the final image tile by tile and output the final image at the exact same resolution as the original resolution.

Thank you so much!

mikebilly avatar Jul 29 '24 18:07 mikebilly