Can Wan3.0's document-to-video fix Sora's drift problem?

alibaba released wan3.0, a video generation model available in beta that creates 30-second clips from text, images, and documents like pdfs and powerpoint files. this update doubles the length of wan2.5 and introduces a multi-modal architecture that processes up to ten images, five videos, and five audio clips in a single prompt to reduce visual drift. pricing starts at $0.05 per second for 480p standard resolution, with a faster prime tier available. alibaba targets sectors requiring high consistency, such as film production, marketing, and generating simulation data for robotics and autonomous vehicles.