The video generation AI 'Wan3.0' has been officially released, capable of generating videos up to 30 seconds long, and boasts significantly improved information analysis accuracy, such as 'creating promotional videos by reading product planning slides.'

Alibaba began general availability of its AI video generation model, 'Wan3.0,' on August 24th. Wan3.0 can read not only text and images, but also documents, spreadsheets, slides, and web pages, and can create videos up to 30 seconds long in a single generation.

Wan 3.0 Video Generation · Cinematic Audio-Visual · Alibaba Cloud Model Studio

This is Wan3.0.
Simple Input. Smart Creation. Where Imagination Meets Reality.
Wan3.0 is now generally available.

🎉 Launch offer: Get 30% off Wan3.0 (Standard) on Alibaba Cloud Model Studio and Qwen Cloud from August 23 to September 23. pic.twitter.com/GVogLtvzjK

— Wan (@Alibaba_Wan) August 24, 2026

Most AI video generation models produce short videos of only a few seconds, and Alibaba's previous model, 'Wan2.7-Video,' could only generate videos up to 15 seconds long at a time. In product introductions and advertisements with narratives, multiple videos are often combined, and the challenge lies in maintaining consistency in people's faces, clothing, backgrounds, and props across scenes.

Wan3.0 extends the maximum video length that can be generated at once to 30 seconds, double the previous version, making it easier to generate a series of scenes including people's movements, conversations, and camera work all at once. It also features a function that suggests the length of the video based on the input instructions, and a video extension function that allows you to create a continuation from an already generated video.

To further reduce the effort required to prepare video materials, Wan3.0 can now directly import office documents. Supported formats include doc, xls, ppt, pdf, txt, key, pages, numbers, and md, and it can also input links to web pages. One file or one link can be imported per request, with a maximum file size of 100MB and a maximum document length of 50 pages.

For example, you can create advertising videos from product description slides or convert training materials into educational videos. It's also possible to visualize numerical data in spreadsheets as dynamic graphs, and it's envisioned for use in creating narrated videos from reports. Essentially, you can use your everyday documents as a starting point for video production.

Wan3.0 allows users to combine multiple materials such as text, images, videos, and audio to use as reference material. Alibaba explains that the function that generates videos based on reference material makes it easier to maintain details such as people's faces, hairstyles, and clothing, as well as props, logos, spatial relationships, and the overall style of the video across different scenes.

Furthermore, the video editing features introduced in Wan2.7 have been carried over to Wan3.0. It allows users to specify and modify the appearance, flow, and dialogue of the video. Alibaba states that they will continue to improve the sound quality and the accuracy of on-screen text.

Wan3.0 is available from Alibaba Cloud Model Studio and Qwen Cloud. The API fees for the Standard version are $0.05 (approximately 8 yen) per second for generated video at 480p, $0.10 (approximately 16 yen) at 720p, and $0.20 (approximately 32 yen) at 1080p. Under normal circumstances, a 30-second 1080p video would cost $6 (approximately 956 yen). As a release campaign, the Standard version will be discounted by 30% until September 23, 2026.

Additionally, a 'Prime' version with improved generation speed is available, with generated videos available at $0.068 (approximately 11 yen) per second for 480p, $0.14 (approximately 22 yen) for 720p, and $0.28 (approximately 45 yen) for 1080p.

Alibaba envisions Wan3.0 being used in films, short dramas, advertisements, marketing, and tourism PR, as well as for creating simulation videos for autonomous driving and robots. Alibaba also stated that it plans to continue improving the accuracy of voice and on-screen text, and develop the Wan series in a direction that allows it to understand richer information and recreate the physical world.