Modelpublic

Raretutor/vdn-minimax-h3-comfyui-int8-convrot

sourceHugging Faceotherupdated 17d agoView on Hugging Face

11likes1.7kdownloads

Model Card

How to use it in ComfyUI ???

<a href="https://www.youtube.com/watch?v=zUk6mqruWdA" style="color: yellow;">https://www.youtube.com/watch?v=zUk6mqruWdA</a>

MiniMax H3 VDN INT8 ConvRot (ComfyUI Version)

This repository provides the INT8 ConvRot quantized linear branch weights for OpenVDN/vdn-minimax-h3.

🚀 Key Improvements:

—~46% Smaller File Size: Reduced from ~4.3 GB down to ~2.10 GB.
—Lower VRAM: Saves ~4.7 GB of peak VRAM during model loading.
—Near-Lossless Quality: Uses Hadamard Convolutive Rotation (ConvRot group 256) to prevent quantization noise while maintaining original BF16 visual accuracy.
—Fully Compatible: Native support in ComfyUI and `RT Minimax H3 VDN`.

📥 How to Use in ComfyUI

Place the downloaded model_int8.safetensors inside your linear branch directory:

text

ComfyUI/models/VDN_Minimax_H3/stage-dmd-step-250/
└── linear_branch/
    ├── config.json
    └── model_int8.safetensors


---

## Credits & License
- Original Base Weights: OpenVDN/vdn-minimax-h3 (https://huggingface.co/OpenVDN/vdn-minimax-h3)
- Project Blog: Video DeltaNet (https://openvdn.github.io/)
- License: MiniMax H3 Community License Agreement (https://huggingface.co/MiniMaxAI/MiniMax-H3)