Krea launches real-time video generation
Krea has released a real-time video tool that generates faster than playback while keeping motion, identity and style consistent, turning AI video generation into an interactive, edit-as-you-go process.
Krea announced Realtime Video on August 27, 2025. The tool generates video faster than it plays back, so users get immediate feedback as they change inputs. According to the company, it keeps output temporally consistent, meaning motion, identity and style hold steady from frame to frame.
Users can drive generation in four ways: painting on a canvas, writing a text prompt, streaming a webcam feed, or streaming their screen. Krea's examples include sketching a scene and watching it animate as strokes are refined, using a webcam as a "rigging tool" with a generative filter to turn a user into a 3D character or a desk into a cityscape, and turning UI motion, cursor moves or timeline scrubs into generative video.
Krea says the approach is "motivated by modern 'world model' ideas," meaning systems that learn how scenes evolve and how actions carry forward in time. The company argues the gain is responsiveness as well as speed. It says even primitive shapes on the canvas can direct complex scenes, which it compares to "video-game-like interactivity."
The announcement leaves out the technical details. It gives no model architecture, resolution, frame rate, latency figures or hardware requirements, and it says nothing about pricing or usage limits. It also doesn't say how temporal consistency was measured or how the system compares with other video models. The post links to a "try now" page, which suggests the tool is available to users, though the post doesn't spell out the terms of access.