pi-H3, a flow-matching video model built on MiniMax-H3, trends on Hugging Face

Lakonik · hf · 2026-11-22

Lakonik's pi-H3 is trending on Hugging Face as an image-text-to-video model. Tags include flow-matching and distillation, with diffusers and ComfyUI support and an linked arXiv paper; the base model is MiniMaxAI/MiniMax-H3. It's a community effort at video/audio generation with distilled acceleration on top of MiniMax-H3 — notable for multimodal practitioners.

Related event: pi-H3 open-sources 2-step video generation, 5s HD in 3.8s on B200(3 posts)→

Original post →

More from Multimodal

Multimodal channel →