MiniMax H3 reenacts movie scenes in a 15-second video, rendered in 322 seconds
miaoying · reddit · 2026-08-04
A Reddit user shows a text-to-video reenactment of movie scenes using MiniMax H3.
- The workflow uses the default T2V setup.
- The clip is 15 seconds long, rendered at 864×480 with 20 steps.
- On a 4090 machine with 64 GB RAM, the render took 322 seconds.
- The author describes the result as an attempt that is also a bit of a blooper, suggesting it is more of a showcase than a polished demo.
Related event: MiniMax H3 Test: 15-Second Video Generation Takes Over 5 Minutes(2 posts)→
More from Multimodal
- A user shares an untested MiniMax I2VA master prompt for local vLLM workflows — Downtown-Cover-7422 · 2026-08-04
- Seedance 2.5 live-performance tests appear to make uploaded audio sound more realistic — AIandDesign · 2026-08-04
- A Reddit clip calls MiniMax H3 “far too fun” — KamuiKakashi · 2026-08-04
- A first text-to-video test runs short at 9 seconds and may need reference images — Sad_Coach_1433 · 2026-08-04
- A duplicate post shares the same MiniMax I2VA prompt for local vLLM 3.5 — Downtown-Cover-7422 · 2026-08-04
- H3 users are struggling to combine custom lip sync with forced start and end frames — chopders · 2026-08-04