"Why do models labeled as 'refiners' always produce repeated images?
#15
by BUCKYGOGO - opened
"Why do models labeled as 'refiners' always produce repeated images?"
This could happen, when the denoising strength is high. Personally i set it to between 0,2 and 0,5 mostly.
Because they're only trained to denoise the final 20% at most, so they only learn micro-details, not composition. Are you trying to produce a whole image with just the refiner?
Try the image dimensions. It's happening because you are most likely asking for a portrait, headshot or closeup and your canvas is tall. Try a square canvas, same prompt. Whatever it is.