Refining distant faces in MiniMax H3: LoRA works up close, wide shots remain hard
robertwellesley · reddit · 2026-10-30
A Reddit thread on optimal workflows for distant face refinement in MiniMax's H3 video model. The author reports Step600 EMA LoRA solves closeup and medium shots, but distant faces remain elusive with sloppy results from most community fixes. They also ask about Star7's one-click face repair ComfyUI node (MiniMaxH3FaceRefineStar7).
More from Multimodal
- Crayon-style cartoon set generated with Midjourney v8.2 and sref 1786649973 — azed_ai · 2026-10-31
- AI-Generated Sitcom Clip Shows Virtual Character Wendy Skateboarding Down a Street — Old_Put1306 · 2026-10-31
- Heavy Wan 2.2 Character LoRA Meets Video: Animate vs VACE vs Scail Workflow Showdown — Business-Chocolate-4 · 2026-10-31
- MOSS AMA Roadmap: Full-Duplex Live Model, Image Turbo, Consumer-GPU Text2Vid — tomByrer · 2026-10-31
- Intel Arc B580/B70 ComfyUI Benchmarks: omnixpu Beats Native pytorchxpu — Willing-Grade5566 · 2026-10-31
- Recreating the Life of King Tut With AI: "The House of Amun" — retarded_raptor · 2026-10-31