MiniMax H3 ref2va struggles at video character replacement, user seeks workflows
alex13331 · reddit · 2026-10-20
Reddit user alex13331 reports that MiniMax H3's ref2va (aka DaSiWa) fails to fully replace characters in video: given a source video for motion and an image for identity, the model keeps too much of the original footage and only swaps parts barely visible in the video. They are asking for prompt or workflow tips, and whether ControlNet is required — highlighting current limits of video character replacement.
More from Multimodal
- DeepSeek 4.1 Flash builds video animation in 12 min for $0.20, self-scores 7.5/10 — oran_ge · 2026-10-20
- Seedance 2.5 to playable game: 3-hour MVP claim from viral skate clip — MoonL88537 · 2026-10-20
- AI Video Name of the Week: Popping Champagne Out of a Moving Mustang, Zero Consequences — mulletproofai · 2026-10-20
- Art Deco Jazz Club prompt template generates striking 1920s illustrations — LudovicCreator · 2026-10-20
- ActionRoPE encodes action as world coordinates for spatially consistent game world models — Zhiyang Deng · 2026-10-20
- UniAVC unifies audio-video generation via progressive training and asymmetric noise scheduling — Kien T. Pham · 2026-10-20