Overview
Tongyi Wanxiang began as Alibaba Cloud’s text-to-image tool for Chinese users, and its portrait and doodle features built a large free user base. In 2026 the main line moved to video. Wan3.0 entered public beta on 6 August and reached general availability on 24 August, extending a single generation to 30 seconds and, beyond text, image, audio and video, accepting documents for the first time, including doc, xls, ppt, pdf and md. At the September Yunqi conference Alibaba showed Wan3.0 in short drama and music video work and named November for the next generation. On 20 October, Wan3.0 Prime took first place on the OpenRouter video leaderboard, with Wan3.0 third. The image side keeps shipping too: multi-image editing, sequential image sets and box-select interactive editing are all available.
Key Features
- 30-second single takes: One generation can run up to 30 seconds, with smart duration suggesting a length from the prompt and video extension continuing a story, leaving room for continuous camera moves and a complete beat.
- Document and multimodal references: Alongside text, image, audio and video, doc, xls, ppt, pdf and md files or links can be supplied as references, at up to 100MB or 50 pages per file and up to 20 reference assets in one call.
- Consistency across a sequence: Characters, props, sets, spatial relations, sound and style hold across a long segment, with faces built around individual bone structure and skin detail so expression and micro-movement track the action.
- Video editing: Visuals, plot and dialogue can be revised on a generated video without redoing the whole segment, with a change confined to the setting you name.
- On-screen text and charts: Long-form text renders across 12 languages, and charts, formulas and infographics can be generated from the prompt, which suits courseware, briefings and product explainers.
- Image-side tools: Multi-image editing fuses up to nine images, sequential storytelling produces up to 12 images with consistent style and subject, interactive editing works from a drawn box, and portrait customisation goes down to bone structure and eye detail.
Use Cases
- Turning an existing product deck or document into a promo or explainer video
- Short drama, music video and advertising teams generating 20 to 30 second continuous segments from a shot list
- E-commerce and brand teams fixing a product and a spokesperson with references to produce a series at volume
- Schools and training teams converting lecture notes and slides into video courses with charts
- On the image side, building poster sets and picture books with nine-image fusion and 12-image sequences
Pros
- A 30-second single take means a narrative does not have to be stitched from short clips
- Documents and web pages can be read, so existing decks and reports become video material
- Per-second billing lets a 480P draft pass and a 1080P finish be budgeted separately
- Editing covers visuals, plot and dialogue, with changes landing locally instead of a full redo
- Reachable through Model Studio, the Wanxiang site and the Qianwen creation apps, with a complete domestic pipeline
Pricing
Wan3.0 is billed per generated second and by resolution. On international channels the standard rate is USD 0.05 per second at 480P, 0.10 at 720P and 0.20 at 1080P; calls through Alibaba Cloud Model Studio in mainland China are priced at RMB 0.3, 0.6 and 1.2 per second for the same tiers. Each platform entry point has its own credits and plans; see the official site for current terms.
Summary
Tongyi Wanxiang has grown from a text-to-image tool into a video-first creation platform. Wan3.0 matters because it lifts the unit of generation from a single shot to a 30-second segment, and because it is willing to read the documents and spreadsheets a team already has. For teams shipping at volume on a budget, per-second pricing plus a 480P draft and 1080P finish split makes trial spending predictable.
Version History
- Wan 3.0 Prime tops the OpenRouter video leaderboard (2026-10-20): Alibaba Tongyi Wanxiang announced on October 20, 2026 that Wan 3.0 Prime ranked first on the OpenRouter video leaderboard, with Wan 3.0 third. Both models entered the new and trending boards at the same time.
- Wan3.0 progress shown at Yunqi, next-gen video model set for November (2026-09-22): At Yunqi 2026 Alibaba demonstrated Wan's video generation in short drama and music video scenarios. Wan3.0 generates a 30-second video in one pass with structured input, and ranks first on both the Artificial Analysis text-to-video and video-editing leaderboards. The next-generation video model is expected in November 2026, targeting longer, more controllable, more complete and more intelligent output, with fine-grained intervention in narrative elements and director-level story understanding and shot organization.