Alibaba Cloud is opening Wan3.0 to public beta, and the headline spec is hard to miss: the video model now generates 30 seconds at a time, up from 15 seconds in Wan2.7 and matching Seedance 2.5.
Wan3.0 doesn't just take text, images, audio, and video prompts. It can also ingest Word, Excel, PowerPoint, PDF, and Markdown documents, then turn them into product demos, course materials, dynamic charts, and business presentations. The model supports smart duration selection and clip extension, and it can follow instructions to tweak the visuals, plot, or dialogue. Reference generation has also improved for character, prop, voice, spatial, and art-style consistency. The polish isn't there yet, though — audio quality and rendering accurate on-screen text are still works in progress.