GitHub radar
MiniMax H3 in ComfyUI: video on your own machine
Comfy-Org packaged MiniMax H3 Turbo for ComfyUI complete with pre-built workflows, three turbo LoRAs, ten style embeddings, and four quantized model variants — 17 million downloads in under a month.
Comfy-Org packaged MiniMax's H3 Turbo video generation model specifically for ComfyUI users, with everything pre-configured. The package includes four quantized model variants (bf16, int8_convrot, fp8_scaled, nvfp4_awq) and pruned options to accommodate different GPU budgets. Three turbo LoRAs enable fast 4-step and 8-step generation instead of the usual 50. Ten specialized embeddings — including fire_breath, dark_magic, bullet_time, and others — can be mixed in to push specific visual styles. Pre-built workflow files cover all three generation modes: text-to-video, image-to-video, and reference-to-video. The int8_convrot variant is recommended with PyTorch cu130.
Why a vibe-coder should care
If you already use ComfyUI, this package gets you started with MiniMax H3 without manually hunting for compatible checkpoints and settings. The turbo LoRAs cut generation time dramatically — 4 steps instead of 50 means near-instant previews. 17 million downloads in under three weeks shows this is where the ComfyUI community landed.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Help me set up MiniMax H3 for ComfyUI from this package: https://huggingface.co/Comfy-Org/MiniMax-H3 — pick the right quantized variant for my GPU, install via ComfyUI Manager and load a workflow for image-to-video generation.
Requires a GPU with 24 GB of VRAM. There's a lighter version — runs on 16 GB.
Open on Hugging Face▌ More finds