ainewsblitz.com

Breaking

Alibaba Launches Wan3.0 for 30-Second Multimodal Video Generation

  • Media Generation
  • Foundation Models

Alibaba has introduced Wan3.0, a video-generation model that can turn text, images, audio, video, documents and web pages into up to 30 seconds of video in a single generation. The model is available through Alibaba Cloud Model Studio in public beta, with API access requiring approval.\n\nWan3.0 expands the Wan family from short-form generation toward longer, more complete scenes. Its reference-based workflow is designed to preserve characters, objects, environments and other visual details while combining multiple input types. Alibaba says the model supports native audio-visual generation, adaptive aspect ratios and output at up to 1080p. The official product page describes the system as an “all-in-one” reference model for directing characters, scenes, camera movement, dialogue and sound through one creative instruction.\n\nThe most notable change is input breadth. According to Alibaba Cloud, Wan3.0 can process documents including PPT, PDF, DOC and XLS files, allowing reports, presentations and other structured material to become video concepts. The API documentation also lists text-to-video, image-to-video, first-and-last-frame generation and reference-based video generation.\n\nWan3.0 supports clips from 2 to 30 seconds, with 480p, 720p and 1080p resolutions. Alibaba Cloud lists pricing at $0.05 per second for 480p, $0.10 for 720p and $0.20 for 1080p, making a 30-second generation approximately $1.50, $3 or $6 respectively. Billing is based on generated duration, and API calls are asynchronous because generation typically takes one to five minutes.\n\nThe model is aimed at filmmaking, advertising, creative design and cultural-tourism workflows, where users often need to combine scripts, reference media and product information. A single continuous generation can reduce the need to stitch together short clips, while native audio generation is intended to keep speech, ambience and on-screen action synchronized.\n\nThe trade-offs are significant. Wan3.0 is currently an application-based public beta rather than a generally unrestricted service, and API access requires approval. Alibaba has not confirmed open-weight availability for the model. Developers also need to use a model, endpoint and API key from the same region, and long-running video requests must be handled asynchronously.\n\nAvailability across third-party platforms remains uneven. The fal Wan3 page describes Wan 3 as coming soon and says Wan 2.7 is currently available there, indicating that Alibaba Cloud Model Studio availability should not automatically be treated as equivalent to broad availability on every hosted inference platform.\n\nWan3.0 therefore represents a shift toward document-aware, multimodal video production rather than a simple increase in clip length. Its practical significance will depend on consistency across complex references, generation reliability, regional access and the quality of the resulting audio-visual synchronization as the beta expands.