Telegram NewsSubmit game
Alibaba's Wan3.0 makes 30-second AI videos from text, images and PDFsImage Source: The-decoder

Alibaba’s Wan3.0 makes 30-second AI videos from text, images and PDFs

George Tsagkarakis Updated 2 min read
Contents 1 sections
We may include affiliate links in our content, meaning we could earn a commission—or receive blockchain-based assets—if you click a link and make a purchase or take a specific action. Additionally, we use generative AI to help draft and refine our posts for clarity and grammar. All content is fact-checked and reviewed by a human editor before publication.

Thirty seconds. That’s the number that matters in Alibaba’s new video model, and it’s exactly double what Wan2.5 could manage.

Wan3.0 is in beta now, and the input list is the stranger part of the pitch. Text works. So do PDFs, web pages and PowerPoint files, which the model chews through and turns into moving footage.

Ten images, five videos, five audio clips, one prompt

Wan3.0 handles text, images, video and audio at the same time. A single prompt can carry up to 10 images, five videos and five audio clips.

That’s a lot of reference material to keep straight, and keeping it straight is the whole problem with generative video. Faces slide. Interfaces smear into nonsense. Props change shape between shots.

Alibaba says Wan3.0 holds detail from reference material more consistently, specifically characters, props and spatial layouts. That’s the claim. Visual drift is the failure mode every model in this category still fights, and a vendor saying it improved consistency isn’t the same as it holding up across a 30-second clip.

It picks the length for you

The model suggests a video length based on your prompt rather than making you guess. There’s also an extension tool for stretching existing videos out further.

Small features, both of them. But if you’ve spent any time generating clips that end mid-gesture, you know why they’re there.

Two tiers, one discount

You can reach Wan3.0 through the wan.video site, Alibaba Cloud Model Studio, or by API on Qwen Cloud. There are two tiers: Standard, currently running a 30 percent discount, and a faster Prime version.

Alibaba is aiming wide with the use cases. Film production speedups. Short dramas. Social clips for creators. Businesses turning text and images into marketing and training video. And developers generating realistic simulation footage to train autonomous vehicles and robotics systems, which is the one use case here that doesn’t need anyone to enjoy watching the output.

The money behind it

Context on the timing: Alibaba is spending hard on AI right now. The company just announced the largest share sale by a Hong Kong-listed company to fund that push.

And last week it reported a 75 percent year-over-year drop in quarterly profit, driven by sharply higher AI investments.

So Wan3.0 arrives with a bill attached. If you’re testing it, start with the Standard tier while that 30 percent discount holds, and throw a document at it before you throw a text prompt at it. The PDF-to-video path is the part no competing model on your shortlist is advertising.

Share this article
George Tsagkarakis

George Tsagkarakis, known as Staycalm4now is a professional author in the crypto gaming industry since early 2018. He has experienced all the growth of Blockchain Gaming and helped multiple projects achieve their goals and established a player base. He is the co-founder of egamers.io and now the Founder and owner of CryptoGames.gg He is also the COO of MyStage, an…

More from AI Video & Media

Subscribe
Notify of
0 Comments
Oldest
Newest Most Voted