Alibaba Unveils AI Video Generation Model Wan3.0, Supporting 30-Second Videos and Multimodal Input
Alibaba launched its AI video generation model Wan3.0 on August 24, enabling single-generation videos up to 30 seconds and supporting multimodal inputs including text, image, video, and audio, as well as document formats. The model also offers API pricing starting at RMB 0.3 per second for 480P resolution, with a 30% discount during a limited promotional period.
Alibaba officially launched its artificial intelligence video generation model Wan3.0 on August 24. The model delivers comprehensive upgrades in generation duration, universal creation, all-around reference, and real-world restoration, capable of generating videos up to 30 seconds in a single run. It also supports document format inputs for the first time, including doc, xls, ppt, pdf, and md. Wan3.0 supports multimodal inputs such as text, image, video, and audio, unifying capabilities like text-to-video, image-to-video, and reference-based video generation.
According to information released by Alibaba Cloud, since its public beta began on August 6, Wan3.0 has demonstrated strong performance in 30-second long video generation, character texture, reference accuracy and consistency, non-realistic stylization, and document-input universal creation. During the beta period, the model was tested in real production workflows, including short drama and film teams producing videos at scale, creator platforms embedding it into their workflows, cultural tourism projects using it to activate landmarks, and musicians using it to create music videos.
In terms of API pricing, Wan3.0 charges RMB 0.3, RMB 0.6, and RMB 1.2 per second for 480P, 720P, and 1080P resolutions, respectively. From August 24 to September 23, Alibaba Cloud's Bailian and Tongyi AI platforms offer a limited-time 30% discount on API usage. Alibaba's previously released quarterly financial report showed that quarterly profit fell by 75% year-on-year, impacted by a surge in AI-related capital expenditure.