The AI video generation landscape has taken a significant leap forward with the release of WAN 2.2, the newest iteration in the WAN series. This open-source model introduces a Mixture-of-Experts (MoE) architecture, boasting 27 billion parameters while activating only 14 billion during inference. This design ensures superior video quality without compromising computational efficiency.

WAN 2.2 offers dual-resolution output, allowing users to generate videos in either 480p or 720p at 24 frames per second. This flexibility caters to diverse project requirements, from quick previews to high-definition presentations. The model excels in producing 5-second videos with intricate details, making it ideal for applications such as character animations, product demonstrations, and creative storytelling.

A standout feature of WAN 2.2 is its cinema-grade aesthetic control. By integrating principles of professional cinematography—including lighting, color, and composition—directly into the generation model, users can achieve precise aesthetic control through multifaceted keywords. This results in visuals that are both stylistically diverse and rich in nuanced detail.

The model's enhanced motion generation capabilities enable precise control and stable generation of complex movements, including nuanced body motion, athletic actions, and detailed facial expressions. The resulting motion is exceptionally fluid, with details rendered in a natural and realistic manner. Additionally, WAN 2.2 features significantly more powerful semantic control and instruction-following capabilities, allowing users to reliably replicate real-world scenarios through text prompts.

WAN 2.2 is available under the Apache 2.0 license, ensuring complete freedom for both research and commercial applications. This open-source commitment invites developers and creators to explore and expand the possibilities of AI-driven video generation.