---

✨ Highlights

• 🧠 Single-Stream Transformer — A unified 15B-parameter, 40-layer Transformer that jointly processes text, video, and audio via self-attention only. No cross-attention, no multi-stream complexity.
• 🎭 Exceptional Human-Centric Quality — Expressive facial performance, natural speech-expression coordination, realistic body motion, and accurate audio-video synchronization.
• 🌍 Multilingual — Supports Chinese (Mandarin & Cantonese), English, Japanese, Korean, German, and French.
• ⚡ Blazing Fast Inference — Generates a 5-second 256p video in 2 seconds and a 5-second 1080p video in 38 seconds on a single H100 GPU.
• 🏆 State-of-the-Art Results — Achieves 80.0% win rate vs Ovi 1.1 and 60.9% vs LTX 2.3 in pairwise human evaluation over 2,000 comparisons.
• 📦 Fully Open Source — We release the complete model stack: base model, distilled model, super-resolution model, and inference code.

🎬 Demo

https://github.com/user-attachments/assets/7050a191-38ef-4e36-8b48-0084ccc694f1

https://github.com/user-attachments/assets/c6cc056f-56ca-4285-80f3-bb6052228d23

🏗️ Architecture

daVinci-MagiHuman uses a single-stream Transformer that takes text tokens, an optional reference image latent, and noisy video and audio tokens as input, and jointly denoises the video and audio within a unified token sequence.

Key design choices:

📊 Performance

Quantitative Quality Benchmark

Human Evaluation (2,000 Pairwise Comparisons)

Inference Speed (5-second video, on a single H100 GPU)

🚀 Efficient Inference Techniques

• ⚡ Latent-Space Super-Resolution — Two-stage pipeline: generate at low resolution, then refine in latent space, avoiding an extra VAE decode-encode round trip.
• 🔄 Turbo VAE Decoder — A lightweight re-trained decoder that substantially reduces decoding overhead.
• 🔧 Full-Graph Compilation — MagiCompiler fuses operators across Transformer layers for ~1.2x speedup.
• 💨 Distillation — DMD-2 distillation enables generation with only 8 denoising steps (no CFG), without sacrificing quality.

📦 Getting Started

Option 1: Docker (Recommended)

# Recommended: use the prebuilt MagiHuman image (supports full pipeline including SR 1080p)
docker pull sandai/magi-human:latest

docker run -it --gpus all --network host --ipc host \
  -v /path/to/repos:/workspace \
  -v /path/to/checkpoints:/models \
  --name my-magi-human \
  sandai/magi-human:latest \
  bash

# Install MagiCompiler
git clone https://github.com/SandAI-org/MagiCompiler.git
cd MagiCompiler
pip install -r requirements.txt
pip install .
cd ..

# Clone daVinci-MagiHuman
git clone https://github.com/GAIR-NLP/daVinci-MagiHuman
cd daVinci-MagiHuman

If you prefer manual setup, follow Option 2 (Conda) below.

Option 2: Conda

# Create environment
conda create -n davinci-magihuman python=3.12
conda activate davinci-magihuman
conda install ffmpeg

# Install PyTorch
pip install torch==2.10.0 torchvision==0.25.0 torchaudio==2.10.0

# Install Flash Attention (Hopper)
git clone https://github.com/Dao-AILab/flash-attention
cd flash-attention/hopper && python setup.py install && cd ../..

# Install MagiCompiler
git clone https://github.com/SandAI-org/MagiCompiler.git
cd MagiCompiler
pip install -r requirements.txt
pip install .
cd ..

# Clone and install daVinci-MagiHuman
git clone https://github.com/GAIR-NLP/daVinci-MagiHuman
cd daVinci-MagiHuman
pip install -r requirements.txt
pip install --no-deps -r requirements-nodeps.txt

# Optional (only for sr-1080p): Install MagiAttention
git clone --recursive https://github.com/SandAI-org/MagiAttention.git
cd MagiAttention
git checkout v1.0.5
git submodule update --init --recursive
pip install -r requirements.txt
pip install --no-build-isolation .

Download Model Checkpoints

Download the complete model stack from HuggingFace and update the paths in the config files under example/.

You will also need the following external models:

🎯 Usage

Before running, update the checkpoint paths in the config files (example/*/config.json) to point to your local model directory.

Note: The first run will be slower due to model compilation and cache warmup. Subsequent runs will match the reported inference speeds.

Input Modes

• T2V — Provide --prompt only and omit --image_path.
• TI2V — Provide both --prompt and --image_path.

Example Scripts

Base Model (256p)

bash example/base/run_T2V.sh   # T2V
bash example/base/run_TI2V.sh  # TI2V

Distilled Model (256p, 8 steps, no CFG)

bash example/distill/run_T2V.sh
bash example/distill/run_TI2V.sh

Super-Resolution to 540p

bash example/sr_540p/run_T2V.sh
bash example/sr_540p/run_TI2V.sh

Super-Resolution to 1080p

bash example/sr_1080p/run_T2V.sh
bash example/sr_1080p/run_TI2V.sh

CLI Mode Selection

• If --image_path is omitted, inference/pipeline/entry.py runs T2V.
• If --image_path is provided, inference/pipeline/entry.py runs TI2V.
• The T2V and TI2V scripts under the same example directory reuse the same checkpoint/config stack. The only difference is whether --image_path is passed.

✍️ Prompt Guidance

daVinci-MagiHuman uses an Enhanced Prompt system that rewrites user inputs into detailed performance directions optimized for avatar-style video generation. For the full system prompt specification, see prompts/enhanced_prompt_design.md.

Below is a quick reference for writing effective prompts.

Output Structure

Every enhanced prompt has three parts:

1. Main Body (150–200 words) — A clinical, chronological description of the character's appearance, facial dynamics, vocal delivery, and static cinematography. Written in English regardless of dialogue language.

2. Dialogue — Repeats all spoken lines in a structured format:

   Dialogue:
   <character description, language>: "Line content"
   

3. Background Sound — Specifies the most prominent ambient sound:

   Background Sound:
   <Description of the background sound>
   

   Use <No prominent background sound> if none.

Quick Example

User input: A man in a yellow shirt says "有的人在一起生活一辈子，还带着假面具呢"

Enhanced prompt (abbreviated):

A young man with short dark hair, wearing a bright yellow polo shirt, sits stationary. His disposition is earnest and slightly agitated... He speaks with a rapid, emphatic tone, his mouth opening wide as he says, "有 的 人 在 一 起 生 活 一 辈 子，还 带 着 假 面 具 呢..." His brow furrows, lip muscles showing distinct dynamics...

Dialogue: <Young man in yellow polo, Mandarin>: "有 的 人 在 一 起 生 活 一 辈 子，还 带 着 假 面 具 呢..."

Background Sound: <No prominent background sound>

🙏 Acknowledgements

We thank the open-source community, and in particular Wan2.2 and Turbo-VAED, for their valuable contributions.

📄 License

This project is released under the Apache License 2.0.

📖 Citation

@article{davinci-magihuman-2026,
  title={Speed by Simplicity: A Single-Stream Architecture for Fast Audio-Video Generative Foundation Model},
  author={SII-GAIR and Sand. ai and Chern, Ethan and Teng, Hansi and Sun, Hanwen and Wang, Hao and Pan, Hong and Jia, Hongyu and Su, Jiadi and Li, Jin and Yu, Junjie and Liu, Lijie and Li, Lingzhi and Ye, Lyumanshan and Hu, Min and Wang, Qiangang and Qi, Quanwei and Chern, Steffi and Bu, Tao and Wang, Taoran and Xu, Teren and Zhang, Tianning and Mi, Tiantian and Xu, Weixian and Zhang, Wenqiang and Zhang, Wentai and Yi, Xianping and Cai, Xiaojie and Kang, Xiaoyang and Ma, Yan and Liu, Yixiu and Zhang, Yunbo and Huang, Yunpeng and Lin, Yutong and Tao, Zewei and Liu, Zhaoliang and Zhang, Zheng and Cen, Zhiyao and Yu, Zhixuan and Wang, Zhongshu and Hu, Zhulin and Zhou, Zijin and Guo, Zinan and Cao, Yue and Liu, Pengfei},
  journal={arXiv preprint arXiv:2603.21986},
  year={2026}
}