// AI NEWS · MODEL RELEASE

Alibaba Launches Wan3.0, an AI Video Model That Makes 30-Second Clips From Documents

Alibaba's newest Tongyi Wanxiang model doubles clip length to about 30 seconds and takes documents — PDFs, slide decks, spreadsheets, web pages — as input, days after a $10 billion share sale to fund its AI push.

2026-08-25 · by Moe Ameen

What happened

Alibaba launched Wan3.0, the newest model in its Tongyi Wanxiang (Wan) AI video line, at a full public release on August 24, 2026, after a public beta earlier in the month. The headline is clip length: Wan3.0 generates videos up to about 30 seconds, roughly double the ~15-second ceiling of the previous Wan 2.x generation, and Alibaba says it keeps character detail, spatial layout, and on-screen motion graphics stable across the full clip rather than degrading after the opening seconds. Reported output runs to 1080p.

The other notable change is what you can feed it. Beyond text and images, Wan3.0 accepts video, audio, and documents — web pages, PDFs, slide decks, and spreadsheets — so a one-pager or a set of slides can become a narrated, animated video. Alibaba has aimed the model at short-drama and film production, advertising and marketing, tourism promotion, and music-video creation, and says it generates multilingual voice and realistic facial expressions in the clip.

Access at launch is hosted through Alibaba Cloud's Model Studio and Qwen platforms, with beta access by application rather than an open consumer app. The release arrived a day after Alibaba raised roughly $10 billion in a share placement earmarked for what it calls full-stack AI — chips, infrastructure, and models. API pricing is usage-based on Model Studio — about $0.05, $0.10, and $0.20 per second of video for 480p, 720p, and 1080p, so a 30-second 1080p clip runs roughly $6 — while open-weight availability was unconfirmed at the time of writing.

Why it matters for creators

  • The clip-length race is real. Thirty seconds is enough for a full hook-and-payoff sequence, not just a loop, which nudges AI video from b-roll toward standalone posts. But a 30-second clip is still one asset, and a feed needs many.
  • Document-to-video lowers the input barrier. If a slide deck or PDF can become a video, the people who never scripted a shoot — marketers, service businesses, educators — can now generate footage from what they already have.
  • Access is gated and cloud-based. Wan3.0 runs through Alibaba Cloud Model Studio by application, so "I saw the demo" and "I can build my week on it" are different things. Plan around availability, not just quality.
  • It is video only. Wan3.0 makes a clip — no captions for the muted feed, no per-platform reframing, no carousel, blog, or newsletter. The finishing and distribution work is still yours.
  • The Chinese-model wave keeps setting the pace. Wan3.0 follows Alibaba's own HappyHorse and ByteDance's 30-second Seedance, so the frontier of length and quality is moving monthly — build on a pipeline that treats the model as swappable.

How to act on this with Kompozy

The useful move today is not to wait for Wan3.0 access — it's to treat any generated clip, from Wan3.0 or the model you already have, as raw material. A 30-second video is long enough to hold three or four distinct moments, and [Kompozy](/) is built to cut and repackage exactly that. Bring the clip in and Kompozy's [Clipped Shorts](/glossary/content-repurposing) split it into separate vertical hooks, burn branded captions for the sound-off feed, and reframe each cut to 9:16, 1:1, or 16:9 per destination — one Wan3.0 generation becoming a handful of posts instead of a single upload.

From there it fans wider than video. The same source feeds a [Carousel](/glossary/hyperframes), quote graphics, a [Persona Short](/glossary/persona-shorts) with a face-locked avatar, a blog article, and an email newsletter, all held to one voice by a [Persona Brief](/glossary/persona-brief) so nothing reads as generic. Then [Autopilot](/glossary/autopilot) schedules and publishes the set across the eight social platforms plus blog and email, with a per-post review gate before anything ships. Because Kompozy treats generators as interchangeable inputs, a model reshuffle next month means you swap the clip, not your pipeline — the same reason you can start with what you have and slot Wan3.0 in the day your access lands.

Quick takeaways

  • Alibaba fully launched Wan3.0 on August 24, 2026, after a public beta earlier that month, reachable through Alibaba Cloud Model Studio and Qwen by application.
  • It generates clips up to ~30 seconds — about double the prior Wan 2.x generation — reportedly at up to 1080p, holding character, layout, and motion graphics steady across the clip.
  • Inputs go beyond text and images to video, audio, and documents (web pages, PDFs, slides, spreadsheets); it also renders multilingual voice and facial expressions.
  • The release followed a ~$10 billion Alibaba share placement to fund full-stack AI; API pricing is usage-based on Model Studio (about $0.05–$0.20 per second by resolution), while open-weight status was unconfirmed at launch.
  • Kompozy turns one 30-second Wan3.0 clip into multiple captioned cuts plus carousels, blogs, and newsletters, then schedules across the eight social platforms plus blog and email.

Frequently asked questions

What is Alibaba Wan3.0?

Wan3.0 is the newest model in Alibaba's Tongyi Wanxiang (Wan) AI video line. It generates clips up to about 30 seconds — roughly double the prior Wan 2.x generation — from text, images, or documents such as PDFs, slide decks, and spreadsheets, and holds character, layout, and motion graphics steady across the full clip. Alibaba fully launched it on August 24, 2026.

When did Alibaba release Wan3.0?

Alibaba rolled Wan3.0 out to a full public launch on August 24, 2026, following a public beta earlier in the month. It arrived a day after the company raised roughly $10 billion in a share placement earmarked for its AI investment.

How long are Wan3.0 videos?

Up to about 30 seconds — roughly double the ~15-second ceiling of the previous Wan generation — reportedly at up to 1080p. Alibaba says the model keeps character detail, spatial layout, and on-screen motion graphics stable across the full length rather than drifting after the opening seconds.

How can creators use Wan3.0 today?

Access is gated through Alibaba Cloud by application, so the practical move is to treat any generated clip as a source. In Kompozy, a single 30-second clip becomes several captioned vertical cuts plus a carousel, quote graphics, a blog, and a newsletter, all scheduled and published across the eight social platforms plus blog and email — swap in a Wan3.0 clip once your access lands.

Related news

← All AI news · Get started →