MiniMax H3 video model demos draw praise for reference-to-video control

On August 16, MiniMax's video generation model H3 dominated community feeds: multiple users posted hands-on r2v / REF2V (reference-to-video) demos, while reviewers offered glowing assessments, noting it is reportedly 33B parameters in size, with strong generation quality, consistency, and controllability, drawing comparisons to an upgraded Sora. The model has already been demonstrated with ComfyUI's rf2va workflow on consumer GPUs, and the community sees it as highly competitive in the video generation space.

Confirmed

Unconfirmed

Why it matters

2026-08-16 ~ 2026-08-16 · 5 related posts

Full story(2 episodes)→

Primary sources