MiniMax-H3 Multimodal Model Surfaces, Runs on ComfyUI with a Single 24GB GPU
reeight · reddit · 2026-08-03
The MiniMax-H3 model (RunningHub version) is now available on Hugging Face. It supports generating video+audio from image/audio/video references (e.g., T2VA, Ref2VA).
According to community feedback, a corresponding ComfyUI custom node repository is already available, with a local deployment threshold of a single 24GB consumer-grade GPU.
Related event: MiniMax Video Model Goes Viral with Stunning User Tests(61 posts)→
More from Multimodal
- Surreal AI-generated landscape hides "another landscape inside the measurements" — piotrbinkowski · 2026-10-16
- Fan animation made in Blender with Claude Code (Opus 5.5) — Icy-Appointment3583 · 2026-10-16
- Making Blender animations with Opus, Astra and Argon: how you prompt still matters — chongdashu · 2026-10-16
- Tessellation-free geometric queries on displaced surfaces cut caustics variance 10x — ssh4net · 2026-10-16
- Complex multi-link CAD mechanism finally animates, dev celebrates the breakthrough — zetalyrae · 2026-10-16
- OpenMOSS releases MOSS-Image-SFT: Apache 2.0 text-to-image model on Hugging Face — bdsqlsz · 2026-10-16