This image shows nine snapshots from Genie 3, Google DeepMind’s new world model. Each scene is an interactive environment, generated from text, navigable in real time. Worlds run at 720p and 24 frames per second, staying coherent for minutes with about a minute of visual memory. You can steer with keys and trigger ‘promptable events’ such as weather shifts or new objects. Compared with Genie 2, the quality and length of interactions are significantly extended. We’re seeing ground-breaking progress with video generation evolving into controllable simulation, opening faster training for agents in synthetic worlds and prototyping for creators and game designers.