Anyone else done some comparisons with refmods vs plain old image refs? I'm seeing worse results?
retained text
I just assumed looking at the demos and the hype that they'll be great and actually started using them. But I then found on a few videos that some clothing details were missing and faces drifted. So I swapped for plan own clothing image reference and face image reference and all of a sudden those details came back. Little things like belt buckles and buttons are now accurate to the reference image whereas they'd get changed using refmods. Anyone else seen this? I don't know why it's worse. (I'm using **Adudeguyman** refmod nodes as he also includes an image in the safetensor)
document fetched 2026-10-29 00:33 · canonical https://reddit.com/r/StableDiffusion/comments/1xhnw3r/anyone_else_done_some_comparisons_with_refmods_vs