Alibaba Launches Wan3.0 Video Generation Model

Alibaba Cloud's English API reference characterizes Wan3.0 as an all-in-one, reference-based video model supporting:

• •Text-to-video generation
• •Image-to-video generation using a first frame or first-and-last frames
• •Reference-based video generation
• •File-based inputs, where the model automatically interprets file contents

The API documentation states that Wan3.0 can generate video up to 30 seconds at 30 frames per second. Generation is asynchronous: developers create a task, receive a task_id, and poll for results. Alibaba Cloud estimates generation tasks typically take one to five minutes, and advises users not to create duplicate tasks while awaiting results.

The service is documented for Alibaba Cloud regions in Singapore, Beijing, and US East (Virginia). Alibaba Cloud also warns that the selected model, endpoint URL, and API key must belong to the same region, or the request will fail.

There is a documentation-status distinction worth noting for teams evaluating availability. The Model Studio launch page describes the Wan3.0 Video API as generally available and callable by developers without an application process. The English API reference, updated August 24, labels the model as being in preview. Teams integrating the service should therefore verify the current status, regional availability, quotas, and billing terms in their own Model Studio account.

Reuters reported that Alibaba said Wan3.0 had been used in short-drama and film production, advertising and marketing, tourism promotion, and music-video creation since its public beta began on August 6. The Next Web reported that the model accepts PDFs and PowerPoint files alongside other media, framing this as a route from structured business material to video output.

For ML and product teams, document and webpage ingestion is a meaningful interface change relative to prompt-only video tools. Systems that turn slides, PDFs, or spreadsheets into video need reliable extraction of layout, tables, images, and semantic hierarchy before generation quality can be assessed. In comparable multimodal-generation workflows, source fidelity, factual grounding, and temporal consistency are distinct evaluation problems: a visually coherent clip can still misrepresent the document from which it was produced.

Alibaba Cloud's launch page advertises time-based billing and a temporary 30% discount on its standard tier through September 24, 2026, with final pricing determined in the console. The page also states that failed generations are not billed. The Next Web reported prices of $0.05 per second at 480p, $0.10 at 720p, and $0.20 at 1080p, while comparing the top-tier rate with Google's Veo 3.1 standard tier. Cross-provider price comparisons require care because resolution, audio features, generation settings, throughput, and output quality vary between services.