Clone the repo:
git clone https://github.com/zai-org/SCAIL.git
cd SCAILInstall dependencies:
# Ensure torch >= 2.4.0
pip install -r requirements.txt| ckpts | Download Link | Notes |
|---|---|---|
| SCAIL-Preview(14B) | 🤗 Hugging Face 🤖 ModelScope |
Supports 512P |
Use the following commands to download the model weights (We have integrated both Wan VAE and T5 modules into this checkpoint for convenience).
# Download the repository (skip automatic LFS file downloads)
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/zai-org/SCAIL-PreviewThe files should be organized like:
SCAIL-Preview/
├── Wan2.1_VAE.pth
├── model
│ ├── 1
│ │ └── mp_rank_00_model_states.pt
│ └── latest
└── umt5-xxl
├── ...
Use the following commands to convert our SAT checkpoint into the .safetensors supported by the Wan2.1 version.
# --scail-dir: path to SCAIL repository folder, default: SCAIL-Preview/
# --save-path: path to save the converted .safetensors file
python convert.py --scail-dir /path/to/SCAIL-Preview/ --save-path /path/to/SCAIL.safetensors Please refer to the guidance in our SCAIL-Pose repository to render a pose video (like rendered.mp4) from a driving video (like driving.mp4).
Use the following commands to run the SCAIL pipeline.
python generate.py --model SCAIL-14B --ckpt_dir /path/to/SCAIL-Preview --scail_path /path/to/SCAIL.safetensors --image examples/SCAIL/ref.jpg --pose examples/SCAIL/rendered.mp4 --prompt "the girl is dancing"Note:
--poseshould be the path to the rendered pose video.
The models in this repository are licensed under the Apache 2.0 License. We claim no rights over the your generated contents, granting you the freedom to use them while ensuring that your usage complies with the provisions of this license. You are fully accountable for your use of the models, which must not involve sharing any content that violates applicable laws, causes harm to individuals or groups, disseminates personal information intended for harm, spreads misinformation, or targets vulnerable populations. For a complete list of restrictions and details regarding your rights, please refer to the full text of the license.
This repository is derived from Wan2.1, which is licensed under the Apache 2.0 License.