MiniMax H3 image-to-video produces heavy artifacts, worse with photos
Radiant-Photograph46 · reddit · 2026-10-27
A Reddit user reports severe artifacts (hands and face) in MiniMax H3 image-to-video. After testing, the workflow isn't to blame: same issue with the default template, base models (fl2v/ref2v), fast h3, and lightx2v LoRA. The determining factor is the reference/first-frame image itself — some images always work well, others always break. Image quality and resolution don't help, and photos trigger it far more often than CGI or 2D art.
More from Multimodal
- Redditor ships ready-to-run Runpod template pairing MOSS-Image with ComfyUI — TheTimster666 · 2026-10-27
- DeepMind's AudioFM hits best results on 10 of 12 audio tasks, beats prior SSL on heart murmur by +4.7 F1 — iScienceLuvr · 2026-10-27
- First 4K tests of Seedance 2.5: stable faces and fine finger motion — DavidmComfort · 2026-10-27
- Seedance 2.5 unlimited generation goes live on Frameo: 3 hours for $19.99 — Aiden_Tech_Ai · 2026-10-27
- A 'Masterpiece' AI Video Made with Seedance 2.5, Prompt Included — Aiden_Tech_Ai · 2026-10-27
- AI cinematic short film 'THROUGH' made with Seedance 2.5 impresses on Reddit — SouthAnimal6134 · 2026-10-27