Identity-aware 4K image editing: a MiniMax H3 single-frame edit workflow
arthan1011 · reddit · 2026-11-04
A Reddit user shares an open MiniMax H3 workflow (on Civitai) exploiting the model's 'identity-aware inpaint' — denoising guided by both text and reference images. Key techniques: upscale references and the single-frame latent to 4MP (8MP for detailed/wide shots), then tiled latent upscale rather than one-shot 4K to avoid style/color shift; every denoise step is grounded by the reference, preserving character identity, which the author claims only video models like Wan2.2 and MiniMax H3 can do well. Also covers prompt style ('actions and results'), turbo LoRA tradeoffs (lightx2v vs taomate), and when face fix is worth it. Multiple example outputs included.
More from coding & agent
- Visual Token Routing cuts VLM agent visual tokens up to 5.8x at same accuracy — wellecks · 2026-11-05
- Model selection is a routing problem: match model size to task ambiguity, says Together — togethercompute · 2026-11-05
- TinyFish becomes Fireworks' web search partner, free live web for open models — pstAsiatech · 2026-11-05
- Perplexity's Computer rebuilds SpaceX 3D engine visualizer from a single YouTube video — AravSrinivas · 2026-11-05
- a16z backs Phaze, Parsec founders' agent-ready remote desktop, in $8.75M seed — a16z · 2026-11-05
- Gemini API's computer use tool is now generally available — patloeber · 2026-11-05