ComfyUI Plugin Adds RefMods: Training-Free References Packed as Safetensors for MiniMax H3
acedelgado · reddit · 2026-09-14
The author updates the open-source ComfyUI-Fantastic-MiniMaxH3-PromptBuilder with RefMods support.
- What RefMods are: reference video/audio assets packed into .safetensors latent format for fast loading, exceeding the native reference limit — a training-free alternative to LoRAs.
- New features: load, create, edit, and use RefMods with a few nodes; a popup library lets you adjust video/audio strengths, enable/disable, and reorder on the fly; the Prompt Builder shows tag previews and a "Draft from RefMods" button auto-fills prompts (e.g., identity RefMods map to <Subject 1> with linked voice files).
- Practical tip: use a Ref2va workflow with community hybrid models, since MiniMax admitted the original open-weights reference model was broken.
Full docs are in the GitHub repo and RefMods Readme.
More from Multimodal
- HyperFrames + HeyGen: decent video from a basic prompt in ~10 minutes — toolstelegraph · 2026-10-30
- Hailuo MCP + Claude: chaining motion graphics generation through an agent — aziz4ai · 2026-10-30
- AI short film Nexus ep 7 cost $28,775 to make, creator shares full breakdown — NightsRadiant · 2026-10-30
- LayerShop: open-source Photoshop-style editor with ComfyUI local AI on layers — HonestTicket6013 · 2026-10-30
- AI video prompt goes viral: rally car racing a collapsing volcanic mountainside — umesh_ai · 2026-10-30
- Redditor finishes a short dinosaur movie made with Veo 3 and Seedance — HeadOpen4823 · 2026-10-30