Alibaba has released the public beta of Wan3.0, a video generation model that doubles the maximum clip length to 30 seconds and accepts text, image, video, audio, web pages, PDFs, and PowerPoint files as simultaneous inputs. The model is available for testing on Alibaba Cloud's Model Studio and Qwen Cloud platforms.

The previous Wan2.7-Video topped out at 15 seconds, a ceiling common among mainstream AI video generators. Wan3.0's native 30-second support enables continuous, unbroken shots and complex camera movements, while an intelligent duration feature recommends optimal length from the prompt and extension tools let creators stretch narrative timelines.

What's new

  • 30-second native clips — twice the duration of Wan2.7-Video and most competing generators.
  • Multimodal reference inputs — text, image, video, audio, web pages, PDFs, and PowerPoint documents processed together.
  • High-precision visual continuity — realistic human faces with synchronized micro-expressions, natural multilingual voice output, stable software UI and motion-graphics rendering.
  • Strict detail replication — character and product details, spatial relationships, and voice consistency preserved from reference inputs.
  • Intelligent duration & extension — model suggests optimal length and can extend generated timelines.
  • Public beta access — apply via Model Studio and Qwen Cloud.

Wan3.0 targets professional workflows across filmmaking, short-form drama, social media, marketing, education, and developer tooling such as simulation data for autonomous vehicles and robotics. The Wan series launched in July 2023 and has iterated toward longer, more controllable, and higher-fidelity output.

Why it matters

Doubling clip length to 30 seconds in a single generation pass reduces the stitching workflow that currently fragments longer AI video projects. Combined with document-to-video conversion, the model lets teams turn slide decks, white papers, and web pages directly into narrated visual explainers — a workflow that previously required separate scripting, asset creation, and editing stages.

Our take

Wan3.0's 30-second ceiling and document ingestion address two practical bottlenecks: clip stitching and the slide-deck-to-video pipeline. The real test will be whether visual consistency holds across the full 30 seconds at usable inference cost, and whether the beta graduates to a priced API with clear credit economics — the wan3.video site already lists Basic, Pro, and Max tiers at $20, $40, and $150 per month.

Sources