MiniMax H3 ecosystem roundup: Character-Swap LoRA final version, retro sci-fi LoRA, ComfyUI Face Refine node
optimisticalish · reddit · 2026-09-29
A Reddit roundup of tools around the MiniMax H3 video model:
- Character-Swap-LoRA (final, 1000 steps): ships with a REF2VA workflow — feed a source video plus a single-character sheet and it swaps the character while keeping everything else intact, transferring the sheet's art style (e.g. comic book). No trigger word needed. Caveat: close-up facial expressions are "often not followed". REF2VA can also do character swapping natively via prompting.
- 1980sscifimovies LoRA: trained on the cyberpunk end of 80s sci-fi, no trigger word required.
- h3-image-edit: a new skill.md letting an LLM write prompts for H3 still-frame anime illustrations covering character/costume changes and minor edits.
- Fizgig 6.3: LoRA trainer update producing "sharper MiniMax H3 photo LoRAs".
- OpenShot 4.0.1: open-source desktop video editor with improved Razor tool and faster color/animation editing.
- ComfyUIMiniMaxH3Director: adds a Face Refine node for ComfyUI with Semantic Bridge semantic enhancement support.
More from coding & agent
- Google Research open-sources PaperOrchestra, a multi-agent framework for writing AI papers — google-research · 2026-11-16
- ffmpeg-skill gives coding agents local video editing via 42 FFmpeg tools and MCP — kajisho5 · 2026-11-16
- Agents Now Reply to Slack Messages on Their Own, Resolving a Prod Incident Ghost-to-Ghost — iamrobotbear · 2026-11-16
- Engineer's Interview Prep Method: Use Claude to Find Which Answers Undersell Your AI Work — brandon_galang · 2026-11-16
- DeepSeek open-sources its app frontend "chat-demo", already at 3.9k stars — NielsRogge · 2026-11-16
- lemire: AI's real impact on coding is expanding the engineer's 'cognitive radius', not speed — lemire · 2026-11-16