Alibaba's Wan adds a realtime text-driven image mode
Alibaba's Wan team launched a Realtime mode on wan.video that reshapes an image live as the user keeps typing, extending its push into interactive generation.
Typing as a live control surface
The Wan team at Alibaba said on Thursday that a new Realtime mode is live on wan.video, pitching it with the line that a user's words become worlds as they write. In the accompanying demonstration, a description typed into the prompt box reshapes the image continuously, sentence by sentence, rather than producing a discrete result after each submission. The team published a demo clip and a direct link for users to try the mode; it did not publish a model card, latency figures or an accompanying technical report alongside the launch.
Part of a wider real-time push
The release lands in a run of interactive-generation work from the same group. Wan-Streamer v0.2, released on 5 July, is an end-to-end duplex model for live audio-visual conversation that responds with synchronized speech and video at roughly 550ms end-to-end latency and around 200ms on the model side, raising output to 640x368 at 25fps. The Wan line more broadly spans text-to-image, image-to-image, text-to-video, image-to-video and editing, with earlier open-weight releases having made it one of the more widely deployed Chinese video generators outside China.
Why it matters
Most commercial image and video tools still work in a submit-and-wait loop: write a prompt, queue a job, evaluate the output, edit, repeat. Collapsing that loop into continuous feedback changes the creative act itself — the prompt becomes a steering wheel rather than an order form — and it changes the serving economics, since a live canvas implies sustained per-user inference rather than discrete billed jobs. It also raises the competitive stakes for rivals whose differentiation rests on output fidelity rather than interaction latency. The open question is quality: real-time modes across the industry have so far traded resolution and coherence for responsiveness, and Alibaba has not yet said where its Realtime mode sits on that curve.