MiniMax H3 long-video coherence caps out around 20 seconds in community testing
Tight_Organization54 · reddit · 2026-09-05
The author polls the community on the longest coherent single-pass MiniMax H3 video and the setups used (model, turbo, sampler, scheduler, steps). Their best result is 20 seconds with the regular flfva model at 8 steps; pushing to 30–45 seconds causes the video to loop the first couple of shots while adding flavors of later ones. They also ask for long-video workflows beyond Continuum, which rarely works for them.
More from Multimodal
- AI video now pulls off orbiting, diving camera moves around a speeding bicycle — umesh_ai · 2026-10-23
- AI art showcase: gothic winged warrior "Azrael-42" with porcelain-crack armor — LudovicCreator · 2026-10-23
- Moody AI-reimagined Harry Potter portraits showcase stunning cinematic generation — azed_ai · 2026-10-23
- Claude Code + ComfyUI MCP workflow for client work: paid models or open weights? — newaccount47 · 2026-10-23
- AI-generated character sheet wows: melancholy traveling cat 'Finch' fully spec'd — azed_ai · 2026-10-23
- Viral prompt template turns AI image models into children's book character-sheet generators — azed_ai · 2026-10-23