Open cilantrocatsoup opened 9 months ago
No errors occurred, but there were two mistakes that have been corrected:
Thank you for your help. The error has been resolved. I feel that disable noise can better blend with the colors of the XL base image, so I did not enable it.
The "add_noise: disable" is used when processing sampling across two KSamplers. It doesn’t make much sense when there’s only one KSampler.
I feel that disable noise can better blend with the colors of the XL base image
However, this is not just your imagination. When sampling without adding noise, the details decrease and it becomes flat. It might be an effective technique to create a gentle atmosphere.
I use KSamplerAdvanced for face replacement, generate a basic image with SDXL, and then use the 1.5 model to redraw the face with Refiner. Three results will emerge:
One is that the face can be replaced normally.
One is that the face is painted with a mask-like appearance.
The other is an error similar to the following: "output with shape [1, 1152, 896] doesn't match the broadcast shape [2, 1152, 896]."
workflow1.json