Tongyi Wanxiang Wan2.1 - An open-source video generation model developed by Alibaba's Tongyi Lab.

## Definition of Wan2.1 Tongyi Wanxiang Wan2.1 is an open-source video generation model developed by Alibaba's Tongyi Lab. It includes four models: two text-to-video (T2V) models (1.3B and 14B parameters) and two image-to-video (I2V) models (both 14B parameters), supporting resolutions of 480P and 720P. The model is designed for high-quality video generation, with capabilities such as complex motion handling, multilingual text effects, and efficient encoding. ## Unique Features of Wan2.1 Wan2.1 offers the following key features: - **High-quality video generation**: Produces realistic visuals with adherence to physical laws. - **Complex motion handling**: Supports smooth and consistent motion in dynamic scenes like sports. - **Multilingual text effects**: Generates advanced Chinese and English text effects for creative applications. - **Efficient encoding**: Uses a custom VAE and DiT architecture, reducing memory usage by 29% and improving speed. - **Physical law simulation**: Accurately simulates collisions, rebounds, and other physical interactions. - **Long-context training**: Ensures high consistency between text prompts and generated videos. ## Unique Features of Wan2.1 In the VBench evaluation, Wan2.1 achieved a total score of 86.22%, ranking first among competing models like OpenAI's Sora, Minimax, Luma, Gen3, and Pika. Its performance highlights its superiority in video generation quality, motion handling, and text-video alignment. ## Model Specifications and Resolutions Wan2.1 includes the following models: - **T2V-14B**: 14B parameters, supports 480P/720P resolutions, optimized for complex motion. - **T2V-1.3B**: 1.3B parameters, supports 480P (720P less stable), runs on consumer-grade GPUs (8.19GB VRAM). - **I2V-14B-720P**: 14B parameters, 720P resolution, for high-resolution image-to-video generation. - **I2V-14B-480P**: 14B parameters, 480P resolution, for efficient image-to-video generation. ## Installation and Usage of Wan2.1 To install and use Wan2.1: 1. **Clone the repository**: `git clone https://github.com/Wan-Video/Wan2.1.git`. 2. **Install dependencies**: Run `pip install -r requirements.txt` (requires torch >= 2.4.0). 3. **Download models**: Obtain models from Hugging Face (e.g., [T2V-14B](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B)) or ModelScope. 4. **Run examples**: - Text-to-video: `python generate.py --task t2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-T2V-14B --prompt "Your prompt here"`. - Image-to-video: `python generate.py --task i2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-I2V-14B-720P --image input_image.JPG --prompt "Your prompt here"`. 5. **Community support**: Join [Discord](https://discord.gg/AKNgpMK4Yj) or the WeChat group for assistance. ## Applications of Wanxiang Wan 2.1 Wan2.1 is suitable for: - **Content creation**: Generating short videos with artistic styles (e.g., oil painting, cyberpunk). - **Advertising**: Designing dynamic ads with personalized text effects. - **Education**: Creating immersive educational videos with dynamic demonstrations. - **Film**: Producing cinematic scenes with complex motion and physics. - **Gaming**: Generating virtual environments for game development. ## Access Points for Wan2.1 Users can access the following resources: - **GitHub repository**: [Wan-Video/Wan2.1](https://github.com/Wan-Video/Wan2.1). - **Model downloads**: [Hugging Face](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B) or [ModelScope](https://www.modelscope.cn/models/Wan-AI/Wan2.1-T2V-14B). - **Official platform**: [Wan Video](https://wan.video/). - **Community support**: [Discord](https://discord.gg/AKNgpMK4Yj) or WeChat group. ### Citation sources: - [Tongyi Wanxiang Wan2.1](https://github.com/Wan-Video/Wan2.1) - Official URL Updated: 2025-04-01