The Wan 2 model family consists of open-weights generative video foundation models developed by Wan-AI in 2025, utilizing a unified diffusion transformer architecture with a custom spatio-temporal variational autoencoder (Wan-VAE). These models support text-to-video, image-to-video, and video editing tasks across various resolutions from 480P to 1080P, with variants ranging from 1.3 billion to 14 billion parameters, all released under the Apache 2.0 license.