HelloPixDig

Directory › Video › DMAD

DMAD
Image: official project repository

> GEM VIDEO · LOCAL AI

DMAD

by Zhengming Yu et al. (Texas A&M, ByteDance)

Generates AI video in 4 steps instead of 50: a MiniMax-H3 version with stereo sound from 8 GB of VRAM, and Wan 2.1 versions under Apache 2.0.

Selon la version : Apache 2.0 (Wan 2.1), licence communautaire MiniMax H3 FreeLocalOpen weights
View on GitHub →

What it does

  • Distillation: 4 compute steps instead of 50 for MiniMax-H3 (1344x768 with stereo audio) and Wan 2.1 (1.3 and 14 billion parameters)
  • 5 seconds from 8 GB of VRAM, 15 seconds with 12 to 16 GB (ready-to-run ComfyUI workflows)
  • Warning: the H3 version needs about 45 GB of system RAM (64 GB advised) and stays under the MiniMax H3 community license

! Not tested by HelloPixDig: figures come from the official Hugging Face page (updated October 10, 2026) and the paper (arXiv 2610.02188). For MiniMax-H3, 5 seconds fit in 8 GB of VRAM but the model is streamed from system memory: about 45 GB are used, 64 GB recommended. DMAD's MiniMax-H3 weights stay under the MiniMax H3 community license, which does not cover the EU, the UK, South Korea or the US without formal authorization from MiniMax. The Wan 2.1 versions (1.3 and 14 billion parameters) are Apache 2.0 but are run with the repository code, with no ComfyUI workflow provided. The SDXL versions are OpenRAIL++-M, and ImageNet-64 is non-commercial.

Install it

Requirements: An NVIDIA GPU (8 GB minimum) and, for the MiniMax-H3 version, 64 GB of RAM advised. ComfyUI or the code repository.

Get the code
git clone https://github.com/Yzmblog/DMAD.git && cd DMAD
Download the weights (Wan 2.1 1.3B example)
hf download ZhengmingYu/DMAD --include "wan2.1/*" --local-dir ckpt
ComfyUI version (MiniMax-H3, 8 GB)
https://huggingface.co/ZhengmingYu/DMAD#ready-to-run-workflows-8-gb-gpu-and-up

The prompt for your agent

Paste it into your coding agent: it checks the requirements and installs it for you.

Claude CodeCodexCursorGemini CLIopencode
Prompt
Install DMAD (https://huggingface.co/ZhengmingYu/DMAD) on my machine.
1. Check the requirements: An NVIDIA GPU (8 GB minimum) and, for the MiniMax-H3 version, 64 GB of RAM advised. ComfyUI or the code repository. If anything is missing, tell me before installing anything.
2. Follow the official method:
- Get the code: git clone https://github.com/Yzmblog/DMAD.git && cd DMAD
- Download the weights (Wan 2.1 1.3B example): hf download ZhengmingYu/DMAD --include "wan2.1/*" --local-dir ckpt
- see https://huggingface.co/ZhengmingYu/DMAD#ready-to-run-workflows-8-gb-gpu-and-up
3. Check that it works, then explain how to launch it and where to start.

Always review what your agent is about to run, especially install commands.