Adobe · 2026-04-10 · major
Adobe MotionStream — Click-and-Drag Real-Time Control Over AI Video Generation, Presented at ICLR 2026
Adobe Research's MotionStream lets users click-drag objects or shift camera angles while AI video is generating — not after. Autoregressive streaming delivers the first segment live while composing the rest. ICLR 2026 paper; public research preview available.
Adobe Research solved one of AI video's biggest unsolved problems: giving creators control during generation, not just after.
What is it?
MotionStream is Adobe Research's experimental AI video generation system that allows users to interact with and direct video while it's being created. Published at ICLR 2026, it's available as a public research preview.
How does it work?
Unlike conventional systems that generate a complete video then deliver it, MotionStream uses an autoregressive backbone where future frames only depend on already-created frames. This enables streaming: the first segment delivers immediately while subsequent sections generate in the background. Users can click objects to direct their movement, adjust camera angles with sliders, and choose which elements stay static — all while generation is in progress. Physics simulation handles secondary effects automatically.
Why does it matter?
Every current text-to-video system produces a fixed output that requires regenerating from scratch if unsatisfied. MotionStream introduces a fundamentally different interaction model — collaborative, iterative generation. ICLR 2026 acceptance signals research credibility, and Adobe's public preview makes it accessible for creative experimentation.
Try it
https://joonghyuk.com/motionstream-web/