Little advice(but important) for minimax h3 r2v users
retained text
if you are using r2v with video refences (for motion etc), dont use any speed lora and make sure you use minimum 20 or high steps. i was trying generate videos for for my dataset sadly i waste alot of time so i dont want you guys to waste your time. resulsts genreated 10 times with 20 steps and 1mp = 9/10 resulsts were perfect besides one. genreated 10 times with 8 steps lora 1mp = 1/10 compelty dogshit results.
document fetched 2026-10-22 00:34 · canonical https://reddit.com/r/StableDiffusion/comments/1xbite5/little_advicebut_important_for_minimax_h3_r2v