Google DeepMind’s Genie 4 creates interactive worlds with sound and talking characters

Google DeepMind has unveiled Genie 4, a world model that generates interactive environments users can explore and change in real time. The research preview adds native audio, more realistic physics, natural human movement and characters that can move, react and speak inside a simulated scene.

Users can begin with text or an image and steer the world while it is running. DeepMind says Gemini models help plan the structure and narrative timeline, while Genie renders the immediate visual and acoustic environment pixel by pixel. Possible uses include game prototyping, robotics training and rapid scenario generation for simulation.

This is still a controlled preview rather than a general release. DeepMind says Genie 4 can lose visual details or object permanence during longer sessions, struggle with complex physical interactions, render inaccurate audio or text, and drift after repeated edits. Generated video and audio will carry Google’s SynthID watermark.

The model is initially available to trusted testers. Google says it will come to AI Ultra subscribers through Project Genie and to selected partners through the Gemini API and Google AI Studio. Genie 4 is a step toward systems that do not only produce video, but maintain a responsive world around a user’s actions.