Alibaba's Video Model Wan3.0 Opens Public Beta: Documents and PPT Can Also Become Video

On August 6, Alibaba opened the public beta of its video generation large model Wan 3.0, upgrading in generation duration, universal creation, omnipotent reference, and real-world restoration. It can generate a 30-second video in a single run, fully expressing creative intent. For the first time, it supports input of document formats beyond the four basic modalities of text, image, audio, and video, including doc, xls, ppt, pdf, and md. The model strives to accurately restore the real world, with each person having a unique appearance and each frame being both realistic and credible. The extension of generation duration is the most intuitive upgrade, allowing complex camera movements like continuous panning and one-shot takes. Wan3.0 also has a smart duration function that automatically recommends an appropriate duration based on the prompt. It is transitioning from a video generation model to an information expression vehicle, capable of reading structured information from documents and PPT, and generating teaching courseware, product demos, and business reports when combined with prompts. The supported formats cover doc, xls, ppt, pdf, md, with individual files or links not exceeding 100MB and 50 pages. For example, uploading a PPT for smart glasses with a prompt yields a complete promotional video. This is underpinned by the model's strong prompt engineering and instruction-following capabilities, which transform prompts into a hub for scheduling input and controlling expression. Video generation models are upgrading from content generation to productivity tools, and users demand not only clarity but also realism. Wan3.0 accurately replicates and presents both real scenes and digital information, with generated human figures aiming for unique faces, detailed facial features and skin, natural and restrained emotional expressions, and coordinated micro-expressions and body movements. In omnipotent reference tasks, fine-grained aspects like character, props, sound, spatial relationships, and style are stably maintained. The visual quality of digital information is also improved, giving charts, data, and interface content a sense of texture. Additionally, Wan3.0's video editing capabilities are greatly enhanced, supporting modifications to visuals, plot, and dialogue. There is still room for improvement in sound quality and text accuracy. Starting immediately, Wan3.0 is in public beta on Alibaba Cloud Bailian, Wanjing Yike, Wanxiang official website, Qianwen Creation PC, IF STUDIO, and Duijiu, with a grayscale rollout on the Qianwen app. API prices for 480P, 720P, and 1080P are 0.3, 0.6, and 1.2 yuan per second, respectively, with the API interface to be fully opened soon.