Alibaba's Wan adds real-time Camera-to-Image generation
Wan launched a Real-time feature suite whose first capability turns a live phone camera feed into stylised imagery, starting with anime looks.
Alibaba's Wan team announced a new Real-time feature suite on Tuesday, led by a capability it calls Camera-to-Image: users open their device camera and see their surroundings restyled on the spot, with the launch demo showing live footage converted into anime-style imagery. The team said the feature is available to try immediately, linking directly to the product rather than a waitlist.
What is different about it
Wan — the international name for Alibaba's Tongyi Wanxiang generative-media line — has until now shipped in the familiar asynchronous shape: submit a prompt or a reference image, wait for a job to finish, collect a file. Camera-to-Image inverts that loop. Instead of authoring a shot and waiting on a render, the user points a camera and the model becomes a filter running against a continuous input stream. That framing puts the feature in the same category as the interactive-latency generation work coming out of Krea, Decart and a handful of research groups, rather than in the text-to-video race Wan is better known for.
Alibaba described Camera-to-Image as the first entry in a broader Real-time suite, implying further capabilities under the same banner. The company has not published latency figures, resolution limits, pricing, or which underlying Wan checkpoint powers the feature, and details available so far come from the team's own announcement.
Why it matters
Real-time generation is the boundary that separates generative media as a production tool from generative media as an interface. Anything running at camera frame rates can plug into live video, streaming, camera apps, virtual try-on and eventually eyewear — surfaces that batch generation cannot reach, and where distribution belongs to platform owners rather than creative-tool startups. Alibaba is one of the few companies holding both a competitive video model family and the consumer apps to put it in front of hundreds of millions of phones.
The cost side is the part to watch. Interactive generation means paying for inference every frame instead of every clip, which forces aggressive distillation and small step counts. If Wan can hold acceptable quality at that budget on consumer hardware or cheap serving, the same techniques transfer straight back to its batch video products — and set a floor that rivals in the AIGC field will have to meet.