Wan
Wan is a family of video generation models from Alibaba’s Wan team, with open-source releases on GitHub and a hosted creative platform now on Wan 3.0.
About this tool
Wan is a series of AI video generation models presented as open and advanced large-scale video models. The open-source Wan2.1 release covers generating video from text or images, editing video, creating images from text and producing audio for video, and includes a small model that runs on consumer graphics cards with around 8 GB of video memory. Models are published on GitHub, Hugging Face and ModelScope.
The hosted Wan platform is meant to make creative work more accessible, offering image and video generation from text or pictures as well as image editing. Its latest version, Wan 3.0, emphasises multi-modal input, cinematic realism and more precise video editing, and lets creators use up to 20 reference assets in a generation. An API and a mobile app are also offered.
What you can do
Open-source models
Download Wan models from GitHub, Hugging Face or ModelScope and run them yourself.
Consumer GPU support
The smaller Wan2.1 model works with around 8 GB of GPU memory.
Multi-task generation
Video from text or images, video editing, image generation and video-to-audio in one family.
Wan 3.0 platform
Hosted creation with multi-modal input and up to 20 reference assets.
Who is it for?
- A researcher fine-tunes an open Wan model for a specialised video task.
- A hobbyist generates short clips locally on a gaming GPU.
- A creator uses many reference images in Wan 3.0 to keep characters consistent.
Before you choose
Running larger open models locally requires substantial GPU memory and technical setup.
Check the licence of each Wan model release and the hosted platform’s terms before commercial use.
Frequently asked questions
Is Wan open source?
Yes. Wan publishes open-source video models such as Wan2.1 on GitHub, Hugging Face and ModelScope, alongside a hosted platform.
Can Wan run on a consumer GPU?
The Wan2.1 text-to-video 1.3B model needs about 8 GB of video memory, so it runs on many consumer graphics cards.
What is new in Wan 3.0?
Wan 3.0 focuses on multi-modal input, cinematic realism and more precise video editing, with up to 20 reference assets per generation.
Check the official website for current features, pricing, and terms.
Similar tools
Seedance
VideoSeedance is ByteDance Seed’s family of AI video generation models, from Seedance 1.0 for multi-shot 1080p video to the newer Seedance 2.x releases.
Qwen Chat
ProductivityQwen Chat is Alibaba’s official assistant for the Qwen model family, offering chat, image and video understanding, image generation, documents and voice.
Stability AI
Image & DesignStability AI develops the Stable Diffusion image models and Stable Audio, offered through an API, open weights and enterprise solutions.
Replicate
CodingReplicate runs open-source and proprietary AI models through a cloud API, and lets developers fine-tune models or deploy their own without managing GPUs.
Hedra
VideoHedra builds visual AI models and infrastructure, offering managed inference for 100+ models, fine-tuning and GPU sandboxes alongside its creative app.
Runway
VideoRunway is a generative AI platform that supports creators and editors in producing text, images, video, and audio.