Announcing Realtime Video

Controllable and frame consistent real-time video generation. Try it out or scroll down to learn more.

Generate faster than playback to get immediate feedback as you change inputs, while preserving temporal consistency so motion, identity, and style remain stable from frame to frame. You can create by painting on a canvas, writing a text prompt, streaming a webcam feed, or streaming your screen.

Behind the scenes, our approach is motivated by modern "world model" ideas: systems that learn how scenes evolve and how actions ripple forward in time. The result is not just speed, but responsiveness with an interactive model that lets you explore ideas without sacrificing coherence. Sketch a scene and watch it animate as you refine strokes. Type a prompt and see the adjustment settle immediately. Pipe in your webcam for live stylization, or point it at your screen to transform UI motions, cursor moves, or timeline scrubs into consistent generative video.

Generating videos this precisely has never been easier. Paint, draw, or move objects on a canvas and let the AI instantly transform simple shapes into anything you can imagine.

Stream your webcam feed and transform your computer camera into a powerful rigging tool with a generative filter. Turn yourself into a 3D character or transform your desk into a cityscape.

Even primitive shapes can act as powerful handles to direct complex scenes. Enjoy video-game-like interactivity for powerful AI video generation.