Alibaba Open-Sources MiniMax-H3 ControlNet-Union with Five Control Types in One Checkpoint

Alibaba's PAI team released and open-sourced MiniMax-H3-Fun-Controlnet-Union on Hugging Face on August 24, providing unified ControlNet control for the MiniMax-H3 video generator. The model is now available for download, with demos also hosted on Hugging Face.

Confirmed

• Trained on the VideoX-Fun pipeline, it is a pipeline-based text-to-video tool officially released by Alibaba PAI
• A single checkpoint of roughly 6.84GB supports five control types: pose, depth, Canny edges, HED edges and MLSD line detection
• Supports Text-to-Video, Video-to-Video, Image-to-Video and other task modes
• Multiple community authors (PurzBeats, linoytsaban, petewoodbridge) covered the release on day one, indicating strong interest in unified ControlNet solutions

Why it matters

Video generation control previously required deploying separate models per control type. By covering five control capabilities with one 7GB unified weight, Alibaba significantly lowers the deployment and usage barrier for controllable video generation with MiniMax-H3.

2026-08-24 ~ 2026-08-25 · 5 related posts

• [source] MiniMax-H3 Fun Controlnet Union model released — physalisx · 2026-08-24
• Alibaba PAI open-sources unified ControlNet for MiniMax-H3: one 7GB checkpoint, five control modes — linoy_tsaban · 2026-08-24
• [source] Alibaba PAI releases MiniMax H3-based Video ControlNet — PurzBeats · 2026-08-24
• Alibaba releases ControlNet-Union for MiniMax H3 on Hugging Face — petewoodbridge · 2026-08-24

1 near-duplicate retellings: alibaba-pai