Wan3.0 Release
According to official announcements from 阿里云, the video generation large model Wan3.0 officially launched today. The model has been comprehensively upgraded in dimensions such as generation duration, universal creation, comprehensive referencing, and real-world scene restoration. It supports generating videos up to 30 seconds long in a single run and adds support for inputting multiple document formats including doc, xls, ppt, pdf, and md, enhancing the fidelity of real-world visuals.
Wan3.0 increases the maximum length of a single video clip to 30 seconds. The longer output duration supports complex cinematic techniques such as continuous camera movement and single-take shots, allowing creators to craft complete narrative arcs and break free from the fragmentation limitations of short clips.
In terms of material input, beyond text, images, audio, and video, Wan3.0 is the first to support multiple office document formats, enabling “video generation from anything.” Supported input formats include doc, xls, ppt, pdf, txt, key, pages, numbers, and md. Users can upload up to one file or link per request, with a file size limit of 100MB and a page limit of 50 pages.
Image realism and content consistency have been heavily optimized. Addressing previous issues of homogenized portraits and distorted textures, Wan3.0's text-to-video portrait module delivers unique facial features for each individual. It finely restores facial features and skin textures, ensuring that micro-expressions, body movements, and emotional expressions align naturally. Even in multi-person scenes, the distinct demeanor of each character is clearly differentiated.
In creative scenarios involving reference image reuse, the model stably preserves all detailed characteristics such as characters, props, spatial layout, and visual style: facial features, hairstyles, and clothing accessories remain consistent throughout; prop structures, logos, and materials show no deviation across different angles; camera positions and character placements maintain logical stability. Meanwhile, it reliably outputs various cinematic styles, such as film and documentary, while minimizing the risk of stylistic inconsistency.
From Wan1.0 to Wan3.0, the Wanxiang model family has undergone a total of 8 iterations. Users can experience it on 阿里云百炼, 万相官网, 万镜一刻, 千问 AI 平台, 千问创作 PC 端, 千问 App, 堆友, and IF STUDIO. Additionally, platforms under 麦芽 such as 跳跃视界, Deevid, and 剧火, as well as 京东灵境平台, 美图, and 井萌, have also completed the integration of the Wan3.0 model.
KEY HIGHLIGHTS
Key highlights
30s Video Creation