Wan
What it is
Wan is a free, open-source suite of video foundation models that runs on Linux and can be self-hosted. It supports text-to-video, image-to-video, video editing, text-to-image, and video-to-audio work. The models can place Chinese and English text inside generated videos. T2V-14B supports 480P and 720P output, while T2V-1.3B supports 480P; the project says 1.3B results at 720P are less stable. The T2V-1.3B model requires 8.19 GB of VRAM. Wan-VAE encodes and decodes 1080P video of any length while preserving temporal information. VACE accepts a prompt with optional video, mask, and image inputs for generation or editing. Model weights are available through Hugging Face and ModelScope. Installation covers pip and Poetry, with PyTorch 2.4.0 or later required. Local Gradio and command-line examples are included. Wan2.1 integrates with Diffusers and ComfyUI, supports commercial use, and uses the Apache-2.0 license.
For and against
For
- Supports text-to-video and image-to-video
- Includes video editing and audio-video tasks
- Runs locally on Linux
- Commercial use is permitted
- Integrates with Diffusers and ComfyUI
Against
- Requires 8.19 GB VRAM for T2V-1.3B
- 1.3B 720P results are less stable
- Maximum listed video length is 0.083333
The verdict
Wan offers a broad, self-hosted model suite for video and related generation tasks. Its hardware requirement and stated resolution limits should guide setup decisions.
Who it suits
Wan suits developers and creators who want self-hosted video generation on Linux. It is also relevant to commercial projects needing text, image, editing, or audio-video tasks.
On the record
- Commercial use
- Yesgithub.com · read 19 Sep 2026
- Downloads
- Model weights are linked for download through Hugging Face and ModelScope.github.com · read 8 Oct 2026
- Generation tasks
- The models support text-to-video, image-to-video, video editing, text-to-image, and video-to-audio tasks.github.com · read 8 Oct 2026
- GPU requirement
- The T2V-1.3B model requires 8.19 GB VRAM according to the repository, which also reports about four minutes to generate a five-second 480P video on an RTX 4090 without quantization.github.com · read 8 Oct 2026
- Hardware requirement
- The T2V-1.3B model requires 8.19 GB of VRAM, and the project says it can generate a five-second 480P video on an RTX 4090 in about four minutes without quantization.github.com · read 8 Oct 2026
- Image and video tasks
- The repository lists image-to-video, video editing, text-to-image, and video-to-audio among its supported tasks.github.com · read 8 Oct 2026
- Image-to-video
- Yesgithub.com · read 19 Sep 2026
- Installation
- The installation guide documents pip and Poetry installation methods.github.com · read 8 Oct 2026
- Integrations
- The project says Wan2.1 T2V and I2V were integrated into Diffusers and that Wan2.1 was integrated into ComfyUI.github.com · read 8 Oct 2026
- License
- The repository identifies its license as Apache-2.0.github.com · read 8 Oct 2026
- Local interface
- The repository includes instructions for running a local Gradio interface.github.com · read 8 Oct 2026
- Maximum resolution
- 720pgithub.com · read 19 Sep 2026
- Maximum video length
- 0.083333github.com · read 19 Sep 2026
- Model downloads
- The project provides model downloads through Hugging Face and ModelScope.github.com · read 8 Oct 2026
- Product
- Wan2.1 is an open suite of video foundation models for video generation.github.com · read 8 Oct 2026
- Prompt extension
- Prompt extension can use the DashScope API or a local Qwen model.github.com · read 8 Oct 2026
- Requirements
- The quickstart says to ensure PyTorch 2.4.0 or later before installing dependencies.github.com · read 8 Oct 2026
- Resolutions
- The repository lists T2V-14B at 480P and 720P, and T2V-1.3B at 480P; it cautions that 1.3B results at 720P are less stable than at 480P.github.com · read 8 Oct 2026
