MiniMax H3 Ecosystem Roundup: Face Refine Workflows, Combat LoRAs, Faster Inference Stack
optimisticalish · reddit · 2026-10-03
Reddit user optimisticalish publishes a daily roundup of MiniMax H3 resources (Oct 3), covering:
- A role-swap workflow with an independent face refine pass: YuNet (ONNX) detects faces, crops a 512 AV latent window, re-samples with a separate int8 model at 12 steps (0.45 denoise), and stitches the crop back into the frame.
- A motion combat reasoning LoRA, seemingly strong for medieval swordfighting.
- A digicam-realism LoRA for amateur footage looks.
- mlsubgen: a multi-language subtitle generator using local transcription plus a local LLM, outputting .SRT files.
- A Claude Code Skill expanding short-code camera prompts (/LOWANGLE /DOLLYIN) into full subject-aware H3 camera prompts.
- A C++/Rust/TypeScript replacement stack to run H3 faster on Intel Arc Pro B70.
The author also indexed all past roundup posts into a themed directory page.
More from Multimodal
- Looped-DiT: 260M looped model beats 6.5x larger ones with 4.9x less inference compute — burny_tech · 2026-10-04
- Augie's AI Image Browser indexes ComfyUI metadata and offers leak-free exports for selling — spanktastic0x · 2026-10-04
- Music cover generated on 8GB VRAM: YuE2 + LTX-2 render in 17 minutes — big-boss_97 · 2026-10-04
- Image-to-image character translation on Runway turns host into an 'Alternative Late Show' — c_valenzuelab · 2026-10-04
- Immunologist generates 2-minute immunology history video entirely in code with Claude — DeryaTR_ · 2026-10-04
- Seedance 2.5's crisp details expose Minimax H3's missing promised 2K update — Cequejedisestvrai · 2026-10-04