Project Genie
Interactive generated worlds
Why it’s worth your time
Project Genie is an early Google Labs research prototype for generating interactive worlds from text or images. A user describes an environment and character, reviews an initial visual sketch, and then enters a navigable version using familiar movement controls. Unlike a conventional generated video, the world continues to form in response to the user's actions. Genie predicts the next visual state while attempting to preserve the immediate scene, character, and direction of travel, creating the sensation of stepping into an image while it is still being imagined.
Existing worlds can be remixed, and uploaded images can seed new ones. That makes Genie a revealing world-model experiment, not a secret replacement for Unity. Controls lag, characters occasionally develop opinions about direction, visual quality can fall under load, and ordinary physics is treated as a polite suggestion. Sessions and memory are short. The prototype is full of seams, but the central trick survives them: generated media responds to movement instead of sitting on a fixed timeline. The path appears because someone chose to walk toward it.
Shared by
Spotted and shared by a Biltib member“Project Genie is rough, delayed, occasionally disobedient - and absolutely a signal. Give it text or an image, step into the result, and the model generates more of the world as you move rather than handing you a finished video. Controls lag, physics wander, and the short sessions make it a research prototype, not a game engine. But the first time the path continues because you walked toward it, the future becomes very easy to see.”
