Alibaba has pushed its Wan video generation line into beta with Wan3.0, a model that doubles output length to 30 seconds and — more interestingly — accepts PDFs, web pages, and PowerPoint decks as direct input. That’s not a minor tweak. Most AI video tools still treat text and images as the ceiling for prompts. Alibaba is betting that the next wave of demand comes from people who already have dense, structured material sitting around and want it turned into something watchable.
A single Wan3.0 prompt can juggle up to ten images, five videos, and five audio clips simultaneously. The model also suggests an optimal video length based on what you’re asking for, and includes an extension tool for stretching existing clips. Alibaba is addressing the persistent visual drift problem head-on — the way faces warp and interfaces distort over longer generations. The company says Wan3.0 keeps reference details like characters, props, and spatial layouts more stable, which matters far more at 30 seconds than it did at 15.
The rollout spans wan.video, Alibaba Cloud Model Studio, and Qwen Cloud’s API. Two tiers are available: Standard, currently discounted 30 percent, and a faster Prime version. Alibaba’s pitch runs the usual gamut — film production, short dramas, marketing videos — but the robotics and autonomous vehicle training angle stands out. Synthetic simulation footage is expensive to produce through traditional pipelines. A model that can generate it from engineering documents could quietly become the more valuable use case than any consumer-facing demo.
The timing is deliberate. Alibaba just executed the largest share sale by a Hong Kong-listed company to fund its AI expansion, and last week reported a 75 percent year-over-year profit drop driven by exactly that spending. Wan3.0 is the product of that bet. Whether the market rewards it depends on whether document-to-video proves genuinely useful or just novel — and right now, nobody outside Alibaba knows which one it is.
