← sources

H3 rev2v sometimes doesnt change the original video at all. Is it just the inference randomness or do i have to do a stronger prompt?

source: reddit (reddit:StableDiffusion)published 2026-11-16 16:30ingested 2026-11-17 02:38

retained text

Im using a simple ref2v of H3 with no loras.

Just replace the character in the video with the character in the image.

But 80% of the time it outputs the same video. Does it need longer prompts, so i get more usable outputs?

document fetched 2026-11-17 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1xxtvtw/h3_rev2v_sometimes_doesnt_change_the_original