Views
No views yet
Qualitative results of video generation using Wan-Alpha. Our model successfully generates various scenes with accurate and clearly rendered transparency. Notably, it can synthesize diverse semi-transparent objects, glowing effects, and fine-grained details such as hair.
| Prompt | Preview Video | Alpha Video |
|---|---|---|
| "Medium shot. A little girl holds a bubble wand and blows out colorful bubbles that float and pop in the air. The background of this video is transparent. Realistic style." |  |  |
1# Clone the project repository
2git clone https://github.com/WeChatCV/Wan-Alpha.git
3cd Wan-Alpha
4
5# Create and activate Conda environment
6conda create -n Wan-Alpha python=3.11 -y
7conda activate Wan-Alpha
8
9# Install dependencies
10pip install -r requirements.txt1torchrun --nproc_per_node=8 --master_port=29501 generate_dora_lightx2v.py --size 832*480\
2 --ckpt_dir "path/to/your/Wan-2.1/Wan2.1-T2V-14B" \
3 --dit_fsdp --t5_fsdp --ulysses_size 8 \
4 --vae_lora_checkpoint "path/to/your/decoder.bin" \
5 --lora_path "path/to/your/epoch-13-1500.safetensors" \
6 --lightx2v_path "path/to/your/lightx2v_T2V_14B_cfg_step_distill_v2_lora_rank64_bf16.safetensors" \
7 --sample_guide_scale 1.0 \
8 --frame_num 81 \
9 --sample_steps 4 \
10 --lora_ratio 1.0 \
11 --lora_prefix "" \
12 --prompt_file ./data/prompt.txt \
13 --output_dir ./outputWan2.1-T2V-14B with --ckpt_dir, LightX2V-T2V-14B with --lightx2v_path, Wan-Alpha-VAE with --vae_lora_checkpoint, and Wan-Alpha-T2V with --lora_path. Finally, you can find the rendered RGBA videos with a checkerboard background and PNG frames at --output_dir.1# An example of prompt.
2This video has a transparent background. Close-up shot. A colorful parrot flying. Realistic style.ComfyUI/models folder and organize them as follows:ComfyUI/models
├── diffusion_models
│ └── wan2.1_t2v_14B_fp16.safetensors
├── loras
│ ├── epoch-13-1500_changed.safetensors
│ └── lightx2v_T2V_14B_cfg_step_distill_v2_lora_rank64_bf16.safetensors
├── text_encoders
│ └── umt5_xxl_fp8_e4m3fn_scaled.safetensors
├── vae
│ ├── wan_alpha_2.1_vae_alpha_channel.safetensors.safetensors
│ └── wan_alpha_2.1_vae_rgb_channel.safetensors.safetensorsComfyUI/custom_nodes folder.
1@misc{dong2025wanalpha,
2 title={Wan-Alpha: High-Quality Text-to-Video Generation with Alpha Channel},
3 author={Haotian Dong and Wenjing Wang and Chen Li and Di Lin},
4 year={2025},
5 eprint={2509.24979},
6 archivePrefix={arXiv},
7 primaryClass={cs.CV},
8 url={https://arxiv.org/abs/2509.24979},
9}