Alibaba’s research unit has released Wan 2.2, an open-source large-scale video generation model distributed under the Apache 2.0 license. The system is built on a Mixture-of-Experts architecture that expands model capacity without increasing compute requirements, making it the first publicly available video generator to employ the technique, according to the developers. The family includes 14-billion-parameter text-to-video and image-to-video models as well as a new 5-billion-parameter model that handles text-plus-image prompts. All versions produce 480p and 720p footage and aim for “cinematic-level” aesthetics while handling more complex scene composition than earlier releases. Pre-trained weights and code are available immediately on Hugging Face, GitHub and Modelscope, enabling researchers and developers to fine-tune or deploy the system without licensing fees. The launch underscores accelerating open-source competition in generative video, two years after the debut of Stable Video Diffusion and similar proprietary tools.