Wan is a family of large generative models for video and images from Alibaba Cloud, with its own creative platform at wan.video. A video generation model is a neural network that creates a video clip from a text prompt or from a starting image. The platform offers text-to-image, image-to-image, text-to-video, image-to-video and image editing.
Some versions are open. Wan2.1 and Wan2.2 are on GitHub under the Apache-2.0 license, with inference code and model weights. Wan2.2 uses a Mixture-of-Experts (MoE) design, in which separate expert networks handle different steps of the denoising process, and its 5B model makes 720P video from text or an image. Newer versions such as Wan2.7 run as an API in Alibaba Cloud Model Studio: a program sends a prompt, and optionally an audio file, and gets back a 720P or 1080P video; without audio, the model adds matching music or sound effects.
The open models are free to download and run. The hosted API is pay per use, priced per second of output video, with a free quota that lasts 90 days.