MiniMax-H3 in Practice: 10s 1MP Clips in Under 6 Minutes on a 5070Ti
Sleepy_Bandit · reddit · 2026-09-18
A Reddit user ran an in-depth hands-on test of MiniMax-H3 for cinematic realism, combat, and scene continuity, and shared a full workflow.
Quality & speed: Higher generation MP drives realism; switching to PlagueKind's workflow plus the fast model minimaxh3fastvideovsadatafree1300step4stepint8convrot lets him produce 1MP, 10-second clips in under 6 minutes at 8 steps on a 5070Ti.
Findings
- Normal motion is solid; fast motion is a weak spot — deroping helps but costs time, frame interpolation smooths jitter nearly free.
- Combat needs a LoRA (combat booster) plus editing.
- Wide shots struggle with faces; FaceRefine helps, best with a single subject.
- Expressions: add a separate expression reference image into retentionanalysis and call it out in the shot description.
- Scene continuity: crop the destroyed-wall shot as a reference, note it in retentionanalysis — damage stays in-scene, occasionally duplicated to the wrong wall. Same trick keeps character positions consistent across shots.
More from Multimodal
- Free course covers the 4 hard problems of production generative media agents — Al_Grigor · 2026-11-04
- PixelBench: testing multimodal models pixel-by-pixel instead of with QA — bdsqlsz · 2026-11-04
- Kitten-TTS 2 offers free voice cloning on laptops, no GPU needed — solyarisoftware · 2026-11-04
- "We'll Take Your RAM": a fully local-hardware cyberpunk meme from 2030 San Francisco — Wonderful-Yak-3348 · 2026-11-04
- Openfilm: open-source AI video agent turns coding agents into video makers — solyarisoftware · 2026-11-04
- Microsoft reopens VibeVoice: 10-second voice cloning, 90-minute local TTS — solyarisoftware · 2026-11-04