Alibaba has launched the beta version of Wan3.0, a generative AI video model that can create videos of up to 30 seconds and process multiple types of reference inputs, including text, images, video, and audio.
The latest model doubles the maximum native video length of Alibaba’s previous Wan2.7-Video model, which supported clips of up to 15 seconds. The longer generation window is designed to enable more complex camera movements, continuous shots, and extended storytelling. Wan3.0 also includes an intelligent duration feature that recommends video lengths based on user prompts, as well as tools for extending generated videos.
A key feature of Wan3.0 is its multimodal input capability. The model can process text, images, videos, and audio, while also supporting web pages and documents such as PDFs and PowerPoint presentations. This allows users to transform information from static or text-heavy sources into dynamic video content.
Alibaba said the model is also designed to improve visual consistency, addressing common problems in AI-generated video such as distorted faces, inconsistent characters, and shifting layouts. Wan3.0 can generate realistic human faces with synchronized micro-expressions, multilingual voice output, stable software interfaces, and motion graphics.
The model provides greater control over reference materials by maintaining details such as character and product appearances, props, spatial relationships, audio, and visual styles. This is intended to help creators produce longer and more coherent scenes while preserving consistency throughout a video.
Beyond creative production, Alibaba sees applications for Wan3.0 across marketing, education, entertainment, filmmaking, and social media. Developers can also use the model to generate realistic simulation videos for applications such as autonomous driving and robotics training.
Wan3.0 is now available for public beta testing, with users able to apply for access through Alibaba Cloud’s Model Studio AI development platform and Qwen Cloud.
Alibaba introduced the Wan visual generation model series in July 2023 and has continued to develop the technology toward more realistic, controllable, and creator-focused image and video generation.
