Xunzhi Xiang, Yabo Chen, Guiyu Zhang, Zhongyu Wang, Zhe Gao, Quanming Xiang, Gonghu Shang, Junqi Liu, Haibin Huang, Yang Gao, Chi Zhang, Qi Fan, Xuelong Li

• Paper release – Publicly available on arXiv ✅ (2025-08-05)
• Demo page release – Launch interactive demo page ✅ (2025-08-05)
• 14B TF Image-to-video inference code release – ✅ (2025-10-21)
• 14B TF Text-to-video inference code release – ✅ (2025-10-21)
• Training code release – Coming soon (ETA: in several weeks)
• Data release – Coming soon (ETA: in several weeks)

💡 Code, models, and dataset will be released in several weeks.

We tested this repo on the following setup:

• Nvidia GPU with at least 32 GB memory for 1.3B models(RTX 4090, A100, and H100 are tested).
• Nvidia GPU with at least 80 GB memory for 14B models(A100, and H100 are tested).
• Linux operating system.
• 64 GB RAM.

Other hardware setup could also work but hasn't been tested.

Create a conda environment and install dependencies:

conda create -n MMPL python=3.10 -y
conda activate MMPL
pip install -r requirements.txt

git clone https://github.com/modelscope/DiffSynth-Studio.git  
cd DiffSynth-Studio
pip install -e .

pip install flash-attn --no-build-isolation

python setup.py develop

huggingface-cli download Wan-AI/Wan2.1-T2V-14B --local-dir .
huggingface-cli download Tele-AI/MMPL --local-dir .

After the downloads complete, your directory layout should be:

├── MMPL_i2v/
│   └── pretrained_models/
│       ├── i2v_14B_6k.pt
├── MMPL_i2v/
│   └── pretrained_models/
│       ├── t2v_14B_8k.pt
├── wan_models/
│   └── Wan2.1-T2V-1.3B
│   └── Wan2.1-T2V-14B
├── README.md
├── demo.png       
└── ...    

Example T2V inference script using the chunk-wise autoregressive checkpoint trained with Teacher-Forcing methods:

cd MMPL_t2v

# Single-GPU, quick validation
bash Wan_t2v_1gpu.bash

# Multi-GPU (4× GPUs), ~20s video with parallel chunking
bash Wan_t2v_4gpu_20s.bash

Example I2V inference script using the chunk-wise autoregressive checkpoint trained with Teacher-Forcing methods:

cd MMPL_i2v

# Single-GPU, quick validation
bash Wan_i2v_1gpu.bash

# Multi-GPU (4× GPUs), ~20s video with parallel chunking
bash Wan_i2v_4gpu_20s.bash

Other config files and corresponding checkpoints can be found in configs folder.

This codebase is built on top of the open-source implementations of:

• CausVid
• Wan2.1
• Self-Forcing

If you find this codebase useful for your research, please kindly cite our paper:

@article{xiang2025macro,
  title={Macro-from-Micro Planning for High-Quality and Parallelized Autoregressive Long Video Generation},
  author={Xiang, Xunzhi and Chen, Yabo and Zhang, Guiyu and Wang, Zhongyu and Gao, Zhe and Xiang, Quanming and Shang, Gonghu and Liu, Junqi and Huang, Haibin and Gao, Yang and others},
  journal={arXiv preprint arXiv:2508.03334},
  year={2025}
}