htdong/Wan-Alpha-I2V

Wan-Alpha++

Animating a Single RGBA Asset into Controllable Transparent Video Effects

Extensive experiments demonstrate that Wan-Alpha++ achieves state-of-the-art performance in image-to-RGBA generation, precise spatial transparency control, and inference efficiency.

---

🔥 News

• [2026.10.19] We release the paper of Wan-Alpha++ on arXiv.
• [2026.10.19] Released checkpoints for normal RGBA video of Wan-Alpha++ on GitHub.

---

📝 To-Do List

• Paper: Available on arXiv.
• Inference Code: Released checkpoints for normal RGBA video of Wan-Alpha++ on GitHub.
• Model Weights of Normal RGBA Video: Released checkpoints for Wan-Alpha++.
• [] Training Code.
• [] Checkpoints for Effect RGBA Video.

🚀 Quick Start

1. Environment Setup

# Clone the project repository
git clone https://github.com/WeChatCV/Wan-Alpha_PlusPlus.git
cd Wan-Alpha++

# Create and activate Conda environment
conda create -n Wan-Alpha++ python=3.11 -y
conda activate Wan-Alpha++

# Install dependencies
pip install -r requirements.txt

2. Model Download

Download Wan2.2-I2V-A14B

Download Wan-Alpha++

Download Wan-Alpha-v2.0-VAE

🧪 Usage

You can test our model through:

torchrun --nproc_per_node=8  generate_2.2_new.py \
        --task i2v-A14B --size 832*480 \
        --ckpt_dir Wan-AI/Wan2.2-I2V-A14B \
        --dit_fsdp --t5_fsdp --ulysses_size 8 \
        --frame_num 49 \
        --vae_lora_checkpoint "VAE/pytorch_model.bin" \
        --high_lora high_noise.safetensors \
        --low_lora low_noise.safetensors \
        --output_dir $output_path/result

You can specify the weights of Wan2.2-I2V-A14B with --ckpt_dir, Wan-Alpha-VAE with --vae_lora_checkpoint, and Wan-Alpha-I2V with --high_lora and --low_lora. Finally, you can find the rendered RGBA videos with a checkerboard background and PNG frames at --output_dir.

Prompt Writing Tip: You need to specify that the background of the video is transparent, the visual style, the shot type (such as close-up, medium shot, wide shot, or extreme close-up), and a description of the main subject. Prompts support both Chinese and English input.

# An example of prompt.
This video has a transparent background. A woman holds a clear cup containing orange juice; she lifts the cup, takes a sip, and smiles.

🤝 Acknowledgements

This project is built upon the following excellent open-source projects:

• DiffSynth-Studio (training/inference framework)
• Wan2.2 (base video generation model)

We sincerely thank the authors and contributors of these projects.

✏ Citation

If you find our work helpful for your research, please consider citing our paper:

@misc{dong2026wanalphaanimatingsinglergba,
      title={Wan-Alpha++: Animating a Single RGBA Asset into Controllable Transparent Video Effects}, 
      author={Haotian Dong and Xin Wang and Wenjing Wang and Chen Li and Jing Lyu and Yan Tina Luximon and Ping Li and Wanchao Su and Ke Xu and Di Lin},
      year={2026},
      eprint={2610.26189},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2610.26189}, 
}

📬 Contact Us

If you have any questions or suggestions, feel free to reach out via GitHub Issues . We look forward to your feedback!