The “Alibaba Cloud” official account announced this evening (the 6th) that the next-generation video generation model Wan3.0 has entered public beta.
Wan3.0 has been comprehensively upgraded in generation duration, versatile creation, comprehensive references, realism, and other dimensions. It can generate 30-second videos in a single run to fully express creative intent. Beyond the four basic modalities of text, images, audio, and video, it supports document inputs such as doc, xls, ppt, pdf, and md for the first time, claiming to make “anything capable of becoming a video.”
Wan3.0 supports generating a single 30-second video, providing greater creative freedom and enabling complex visual language such as continuous camera movement and uninterrupted single-take shots to be fully expressed, moving from “generating a shot” to “telling a complete story.”
Wan3.0 also supports an intelligent duration feature that can automatically recommend a suitable duration based on the prompt. Combined with the video extension feature, it can extend the story and plot, allowing ideas to develop naturally.
Wan3.0 supports comprehensive references across text, images, audio, and video, and can also read structured information from documents, spreadsheets, webpages, and more. Combined with prompts, it can generate teaching materials, product demonstrations, animated charts, business presentations, and other content, meaning that large amounts of office material can be directly converted into video.
Supported input formats include doc, xls, ppt, pdf, txt, key, pages, numbers, and md. A maximum of one file or link can be entered, with a file size of no more than 100MB and no more than 50 pages.
To move beyond the greasy and uniform appearance of AI-generated characters, Wan3.0’s people in text-to-video mode aim to be “different for every person”: facial features and skin details are rendered more sensitively, character emotions are expressed naturally and with restraint, and micro-expressions coordinate with body movements. Even in group scenes, it can delicately portray the complex emotions of different characters.
For comprehensive reference tasks, Wan3.0 can accurately reproduce reference details while maintaining consistency in key dimensions such as characters, props, sound, spatial relationships, and style. For characters, fine-grained features including facial features, hairstyle and hair color, body shape, clothing, and accessories can all be stably preserved, avoiding visual discontinuity. For props, their appearance from multiple angles, hardware structure, logos, and material details remain consistent. For scenes, it accurately handles the relationship between character positions and camera viewpoints. For style, it precisely renders the tone of film and television production, while also reducing style confusion in multi-style creations.
Starting today, Wan3.0 is available in public beta on Alibaba Cloud Bailian, Wanjing Yike, the Wanxiang official website, and the Qianwen Creation PC client, with a phased rollout in the Qianwen app. API prices for 480P/720P/1080P are respectively 0.3/0.6/1.2 yuan per second, and the API will be fully available soon.
Try it here:
Alibaba Cloud Bailian: https://bailian.console.aliyun.com/cn-beijing ?tab=model# /model-market/detail/wan3.0-video
Wanjing Yike: https://www.yikeai.com/
Wanxiang official website: http://wanxiang.aliyun.com
Qianwen Creation PC client: https://c.qianwen.com/
IF STUDIO: https://ifs.iconfont.cn/
Duiyou: https://d.design/