Event date · · MiniMax

Alibaba PAI released MiniMax-H3-Fun-Controlnet-Union-2.0, a ControlNet-Union branch for MiniMax-H3 supporting 8 control conditions and video inpainting

MiniMax 稀宇科技Chinese AIOpen weights
FACT STATEMENT

Alibaba PAI released MiniMax-H3-Fun-Controlnet-Union-2.0 on Hugging Face, a ControlNet-Union branch for MiniMax-H3 supporting 8 control conditions and video inpainting. The checkpoint is about 13.5 GB and requires minimax_h3_control_inpaint_post_norm.yaml.

China context

Outside China
Open weights · huggingface.co
Claims
Company-reported; not yet independently evaluated
For builders
Builders outside China can download the checkpoint from Hugging Face and integrate it with MiniMax-H3 for controllable video generation and inpainting.
For investors
Investors should note Alibaba PAI's continued investment in open-weights video generation models, which may pressure competitors and expand the developer ecosystem.
What happened

Alibaba PAI released MiniMax-H3-Fun-Controlnet-Union-2.0 on Hugging Face. This model is a ControlNet-Union branch for MiniMax-H3, supporting 8 control conditions (Canny, Depth, HED, MLSD, Pose, Scribble, Layout, Gray) and video inpainting. It uses a denser control injection with 10 control blocks across 50 transformer blocks, guidance-distilled for guidance scale 1.0, and a post-norm inpainting recipe. The checkpoint is about 13.5 GB and must be used with the minimax_h3_control_inpaint_post_norm.yaml config.

Technical significance

The model expands control conditions from 5 to 8, adding Scribble, Layout, and Gray. It doubles control injection points from 5 to 10 blocks, improving structural adherence. Inpainting uses a post-norm recipe where masked pixels are zeroed after ImageNet normalization, improving blending. The control branch is guidance-distilled, requiring guidance scale 1.0 and one forward pass per step.

Industry impact

Developers outside China can now use a single checkpoint for 8 control conditions and video inpainting with MiniMax-H3, reducing the need for multiple specialized models and simplifying video-to-video generation workflows.

Decision value

This release lowers the barrier for developers to implement controllable video generation and inpainting with MiniMax-H3, potentially reducing development time and compute costs by avoiding per-condition model switching.

What to watch

Next signals to check: whether Alibaba PAI releases a full MiniMax-H3 base model with this control branch integrated, and whether the model is adopted in open-source video generation pipelines.

Latest in Chinese AI

  1. Manus AIManus AI Launches Video Editor in Manus 2.0
  2. DeepSeekDeepSeek open-sources dsh-libreoffice-kit 0.1.0 for font-friendly Office conversion and rendering in Node.js
  3. DeepSeekDeepSeek open-sources DeepEP-Ascend and DeepGEMM-Ascend for Huawei Ascend NPUs
  4. Beijing Academy of Artificial IntelligenceBAAI released AREX-2, a 27B self-improving agent model, on Hugging Face
  5. Shanghai AI LaboratoryShanghai AI Laboratory open-sources AdvancedMathBench for proof generation and verification

All China AI Events

AIGC Newsletter

China AI, with sources and context.

Analysis of Chinese AI models, companies and policy, and what you can use outside China.