MiniMax Updates Community Catalog for #MiniMaxH3

MiniMax updated the official community catalog awesome-minimax-h3-integration for the H3 video model, adding the 4-step Sol-H3 distillate, FastH3-Omni 8-Step, and the Ref2VA-VSA ComfyUI node, along with quantizations for local execution on 8-24 GB VRAM.

Compiled by Sergey KostenchukPublished 2026-10-26Updated 2026-10-26

2026-10-26 Coding HuggingFace

🛠 MiniMax Updates Community Catalog for #MiniMaxH3

The official catalog awesome-minimax-h3-integration (commit 25.10.2026) now features FastH3-Omni 8-Step V1 — video with stereo sound based on up to 12 references, the 4-step Sol-H3 distillate using sparse attention SOL/BSA, and the Ref2VA-VSA ComfyUI node.

🌍 The community is systematically compressing generation: 4-8 diffusion steps instead of dozens, sparse attention reduces inference costs, and quantizations allow the model with 464 GB of weights to run on cards with 8-24 GB VRAM — changing the economics of video production and API services.

👤 Local execution has become a reality: on 24 GB VRAM — pruned int8 at 19.5 GB, on 12-16 GB — GGUF Q4_K_M, on 8 GB — via DiffSynth-Studio. Weights, quantizations, and ComfyUI nodes are collected in one catalog; verify speed measurements on your own hardware.

Source 1: https://github.com/MiniMax-AI/awesome-minimax-h3-integration Source 2: https://huggingface.co/LiconStudio/Minimax-H3-VBVR

Sources