Trained to Imagine
Rendering
Creative Intelligence
AI Video / Image Models

Alibaba unveils Wan3.0 with 30-second clips and mixed inputs

Source: Alibaba Cloud ·

Wan3.0, released in beta, doubles maximum clip length to 30 seconds and accepts text, image, video, and audio inputs together, alongside documents such as PDFs and slides. It can also suggest a clip length from the prompt and extend existing videos.

TTI Takeaway

Thirty seconds is a full spot, not a shot. Accepting a deck or a PDF as input is aimed squarely at marketing teams generating from the brief itself — which makes the brief, not the prompt, the thing that needs to be well written.

Analysis

The more telling part of the Wan3.0 announcement is not the length but the input list. Alibaba describes a model that takes text, images, video and audio together, and also reads web pages, PDFs and PowerPoint files. That is a model designed to sit at the start of a marketing workflow rather than at the end of a director’s shot list, and it says a lot about which customer Alibaba expects to pay for longer generation first.

The specifics, from Alibaba’s own post: Wan3.0 is in public beta, with access by application through Alibaba Cloud’s Model Studio and Qwen Cloud. Native clip length rises to 30 seconds from the 15-second ceiling of its predecessor, Wan2.7-Video. An “intelligent duration” feature proposes a length based on the prompt, and an extension feature lengthens existing clips. Alibaba also claims realistic faces with synchronised micro-expressions, multilingual voice output, stable rendering of software interfaces and motion graphics, and reference fidelity for characters, props, audio, spatial layouts and styles, including what it calls strict control over character and product details. The listed uses run from filmmaking and short dramas to marketing, education and simulation footage for training self-driving and robotics systems.

For a commercial production, the product-detail claim matters more than the duration. A 30-second spot that drifts on the label, the logo or the packaging is unusable no matter how fluid the motion is, and the agency will discover that in client review, not in the prompt window. If reference fidelity holds up, Wan3.0 becomes plausible for animatics and pitch films that need to show the actual product rather than a lookalike. If it does not, the extra fifteen seconds simply give the drift more time to accumulate. The way to find out is unglamorous and specific: feed it the real packshots and brand guidelines and compare the last second of the clip against the first.

Document input changes who is effectively directing the model. When a brief or a deck becomes the prompt, the person writing the brief is writing direction, often without knowing it. A contradictory deck, with a mood page that says warm and handheld and a strategy page that implies cool and locked off, will produce contradictory output. We think production teams will end up reviewing briefs for how well they generate, the way they currently review scripts for budget, and that is a skill agencies should build on purpose rather than discover in a failed round.

Several things the post does not state: output resolution, pricing, and any date for general availability. Those decide whether this is a delivery tool or a previs tool. There is also the provenance question for European campaigns. The EU AI Act labelling rules that took effect on August 2 put marking duties on providers, and any agency running Wan output in Europe should ask how, and whether, generated files are marked before the work goes out. With Black Forest Labs’ FLUX 3 Video arriving the same week at a 20-second ceiling, the length race is clearly on; the more useful comparison will be which model still holds a product shot steady at the end of the clip.