MOSS and same-image consistency
retained text
How to keep the same face/clothing on a subject that appears twice in the same image? Qwen 2.1 is stellar at it, why isn't Moss? Specifically prompting for a split image where a person appears on the left side and also the right side, but in different contexts. It's manageable when the subject is in the same pose, it completely falls apart when you change pose or setting. The identity is lost and Moss generates two different people. Am I doing something wrong? Or is Mosss simply not good at it? I'm using Kromoss (built on Moss-base) and the 8 steps Lora.
document fetched 2026-10-21 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1xax3y9/moss_and_sameimage_consistency