Wan3.0 expands video generation to document inputs and longer clips

Documents join the reference inputs

Alibaba's Wan3.0 extends its video-generation service to document inputs and longer single generations. The company's August product announcement describes up to 30 seconds of output with native audio and resolutions up to 1080P. Text, images, audio and video can be supplied as references, alongside documents or web material.

The document feature changes what can serve as the source of a video. A presentation or report can supply content directly rather than requiring the user to rewrite every point into a text prompt. It also adds a fidelity problem: a generated film must preserve the meaning of the source material as well as produce convincing motion.

The published input limits

Alibaba Cloud's detailed guide lists Word documents, spreadsheets, presentations, PDFs, plain text, Markdown and Apple's office formats among the supported inputs. A request can contain one document file or link, no larger than 100 MB and no longer than 50 pages.

The guide also describes automatic duration suggestions, video extension and revisions to visuals, plot or dialogue. These features address different stages of production. Duration selection sets the initial canvas; extension continues an existing sequence; editing changes material already generated. A 30-second maximum does not mean every request must produce a full half-minute.

Reference control and stated limitations

The product announcement specifies up to ten image references, five video references and five audio references. Alibaba positions these inputs as ways to preserve characters, objects and style through a generated sequence, while the detailed guide emphasizes more natural faces and consistency between references and output.

The same guide acknowledges that audio texture and the accuracy of text rendered inside a video remain areas for improvement. That qualification is particularly relevant to document-driven output. A model can communicate the general story of a report while still rendering a label or number incorrectly; the input capability and exact reproduction of every source detail are separate questions.

Budgeting for revisions as well as finished seconds

The price schedule makes experimentation a measurable part of a production budget. Four 30-second drafts at 480P would cost $6 at the published rate, before generating a final 1080P version for another $6. That $12 illustration assumes all five generations are billable at the stated rates; it does not include unrelated editing services or promise that one final pass will be sufficient.

Alibaba describes the editing capability as carried forward from Wan2.7, while documents are a new input for Wan3.0. That separates a new source-material route from an existing revision tool. The available product pages establish generated-video and editing functions, but do not promise an editable timeline, separate actor layers or a scene file suitable for a conventional 3D production package.

How the service is priced

The standard international rates listed by Alibaba Cloud are $0.05 per generated second at 480P, $0.10 at 720P and $0.20 at 1080P. At those rates, a 30-second output costs $1.50, $3 or $6 respectively. These calculations describe one generation and do not include the cost of creating alternatives.

The August product page also gives yuan-denominated regional prices and says failed generation requests are not charged. Those terms should be read within their service region, rather than converted into an assumption that every account receives the same billing arrangement.

Wan3.0 is available through Alibaba Cloud Model Studio. The evidence here establishes a hosted model release with documented inputs and commercial access. It does not establish an open-weight release, even though earlier members of the wider Wan family have been distributed through other channels.