Alibaba Releases Wan3.0 to Generate 30-Second AI Videos

Alibaba has launched the beta version of its latest video generation model, Wan3.0, introducing significant upgrades in multimodal input capabilities and output length. The new model can generate videos up to 30 seconds long, doubling the capacity of the previous Wan2.5 version. Notably, Wan3.0 can process text, images, video, and audio simultaneously. A single prompt can ingest up to ten images, five videos, and five audio clips. It also supports static documents as inputs, allowing users to convert PDFs, web pages, and PowerPoint presentations directly into dynamic video content.

To address common issues in AI video generation, such as facial distortion and visual drift, Alibaba designed Wan3.0 to maintain high consistency for characters, props, and spatial layouts. The system also features an extension tool to lengthen existing video clips and can automatically recommend the optimal video length based on the user's prompt. Practitioners can access the model in two tiers: a Standard version, which is currently offered at a 30 percent discount, and a faster Prime version. The model is available through the wan.video website, the Alibaba Cloud Model Studio, and via an API on Qwen Cloud.

For developers and creators, these features open up new workflows. Film producers and social media creators can accelerate editing, while businesses can automate the creation of marketing and training materials from existing documents. In technical fields, developers can leverage the model to generate realistic simulation footage for training robotics and autonomous vehicles. This release comes as Alibaba aggressively expands its AI footprint. The company recently initiated the largest share sale by a Hong Kong-listed firm to fund its AI initiatives, despite reporting a 75 percent year-over-year decline in quarterly profit driven by these heavy capital investments.