Alibaba Wan3.0 Generates 30-Second Videos From Documents

Alibaba has opened public beta access to Wan3.0, its latest video generation model, on Qwen Cloud and Alibaba Cloud Model Studio. The model represents a significant leap forward from its predecessor, Wan 2.7, which was limited to generating clips of 2 to 15 seconds. By natively producing 30-second video clips in a single generation pass, Wan3.0 allows creators to build full broadcast-length spots without the tedious process of stitching multiple short clips together.

A standout feature of the new model is its Omni-Reference system. While traditional video generators only accept text, images, or audio prompts, Wan3.0 can ingest structured documents. Users can feed the model files in formats such as .doc, .xls, .ppt, .pdf, .txt, .key, .pages, and .numbers, as well as live URLs. This capability enables practitioners to turn spreadsheets, presentations, and web pages directly into video content, streamlining the pre-production workflow.

Under the hood, Wan3.0 is built on a Diffusion Transformer architecture. It features production-grade character consistency and in-pass audio generation. Alibaba has also detailed its API pricing structure ahead of a full rollout. Generating video will cost $0.05 per second at 480p resolution, $0.10 per second at 720p, and $0.20 per second at 1080p. Under this pricing model, a complete 30-second high-definition clip will cost $6.

The AI development community has reacted with cautious optimism. While the technical capabilities are impressive, developers are waiting to see if Alibaba will release the model's open weights under an Apache 2.0 license. The company has a history of unfulfilled open-source promises, leaving some practitioners hesitant to fully integrate the technology until the licensing terms are finalized.