MiniMax H3's Reference Images Win Over an Open-Weights-Only Video Generation Holdout
Peregrine2976 · reddit · 2026-09-03
An open-weights-only creator says MiniMax H3 finally got him into video generation: Wan 2.2 never hit the quality or consistency he wanted, and the tools that did were closed-weight paid products.
- MMH3 is multimodal with vision, so you can feed reference images/character sheets directly — the feedback loop for teaching new concepts is far shorter than training LoRAs
- He doubts character/clothing/setting LoRAs will go obsolete, but expects far fewer will be needed
- Full workflow (Pastebin) and the reference sheet used (Imgur) are shared for reproduction
His stated motivation is ideological: tech that can be hacked apart and isn't beholden to anyone — same reason he runs Linux and Firefox.
More from Multimodal
- Luma shares filmmaking-style workflow: motion-capture real stunts then restyle with video-to-video — lmoroney · 2026-10-18
- MOSS-Image-SFT at 4 Steps Draws Hype: 320x320 to 2K Without Artifacts — Powerful_Evening5495 · 2026-10-18
- MOSS SFT vs Flux 3 Klein vs SD 3.0: Same-Prompt Comparison Exposes SD3's Age — ZootAllures9111 · 2026-10-18
- Dev builds Mumbai-inspired 3D manga background generator from OpenStreetMap data — CurieuxExplorer · 2026-10-18
- Omni 1'1 flash video demo generates atmospheric foggy Victorian haunted house — Dense-Instruction210 · 2026-10-18
- Flux 3 klein distilled tested: int8 workflow delivers surprisingly good 2MP images — Ok_Tale7582 · 2026-10-18