Alibaba doubled the clip length to 30 seconds. The length is not the point.
You make short videos. The annoying part? A single shot gets cut off at 15 seconds, so telling a complete story means stitching seven or eight clips together, and the rhythm falls apart. On August 24, Alibaba shipped Wan3.0, which generates up to 30 seconds in one pass, double the 15-second cap of its predecessor Wan2.7.
The length is just the appetizer. In 2026, 30 seconds is no longer exclusive: ByteDance's Seedance 2.5 global edition also does 30 seconds, and Google's Veo 3.1 does 20. So "half a minute of generation" is becoming a table stake, not a moat.
What actually lands: drop a document in, get a video out
Wan3.0 is the first in the line to accept office documents as input: doc, xls, ppt, pdf, md, even a web link. Hand it a product deck and it returns a narrated video. For marketers and internal comms teams, that beats writing a prompt that describes what you want.
Quality reaches 1080p. Micro-expressions and the distinct faces of multiple people on screen are improved. Alibaba says it fixed the old problems of homogenized faces and plastic texture. Continuous camera moves and single-take shots now hold up. From Wan1.0 to 3.0, the family has iterated eight versions, and Jumping Vision, JD Lingjing, and Meitu have already integrated it, faster ecosystem spread than the last generation.
A concrete scenario
You run a Taobao store and want a 15-second product video for a new launch. The old paths: write a long prompt and wrestle with the AI, or pay a few hundred yuan for an agency. Now you drop the product doc and a few images into Wan3.0 and it returns a narrated draft. Two tweaks and you ship. Agency fee gone, timeline from two days to two hours.
This is exactly the capability small merchants lack most. Not "knowing how to write prompts," but "turning documents you already have into video."
The price, translated
Top tier is $0.20 per second at 1080p, $12 a minute. Sounds steep? Google's Veo 3.1 charges $0.40 per second on standard tier, $24 a minute. For the same minute of footage, Wan3.0 runs about half the price of Veo 3.1.
480p is just $0.05 per second, 720p $0.10. Plenty for a cheap first pass. Against per-clip agency fees of hundreds of yuan, the marginal cost of self-serve generation is nearly nothing.
Two cold showers
First, the weights are closed. The Wan line built its reputation on open source. Wan3.0 shuts the weights and moves to an application system, on Alibaba Cloud's Model Studio and Qwen Cloud, while the consumer site wan.video is members-only. The commercial instinct is strong, a sharp turn from its open-source persona.
Second, no independent benchmark yet. Every claim above about quality and consistency comes from Alibaba itself. Third-party blind tests have not landed. Do not trust the launch copy blind.
One more worry: doc-to-video is garbage-in, garbage-out. If your deck is messy, the video will not save it. Do not expect it to figure out what you meant to say. Right now it fits internal drafts best, with a human review before anything goes live.
Why now
The same week, Alibaba raised $10.2B in a Hong Kong share placement and reported a 75% drop in quarterly net profit, all of it poured into AI infrastructure. Cloud and AI revenue rose 45% year over year to 48.4B yuan, but capital expenditure hit 67.6B yuan. It is more eager than anyone to make AI pay back, and that pressure lands on pricing and experience. So cheap today is not a promise of cheap forever.
Why this matters to you
If you run e-commerce, a channel, or internal comms, Wan3.0 is worth a look. Turning documents into video cuts a real chunk of editing outsourced work. But decide first whether you need "a video exists" or "a video you can ship." Right now it fits internal drafts best, with a human review before launch. Tight budget? Start at 480p, do not jump to 1080p. And do not let the "30 seconds" headline distract you. That is now industry standard. The part worth money is the document-in, video-out pipeline.