Google DeepMind debuts Genie 4 world model with native audio

Genie 4 is Google DeepMind's first real-time world model to generate audio natively, the lab announced, with music, ambient noise and interactive dialogue produced simultaneously with the simulated scene.

Testers can also steer the world as it unfolds. The post says users can 'use natural language to modify a simulation in real-time — like changing the weather, moving the camera, or introducing a new person or object,' and that characters 'converse and react to your inputs or events in the world, all in real-time.'

Behind the scenes, DeepMind has wired its own Gemini stack into the loop. 'Gemini Omni anchors the simulations in real-world appearance and behavior, while other Gemini models provide the complex reasoning needed to help maintain a simulation's coherence and continuity,' the authors write. A smaller model does the frame-by-frame work: 'Gemini 3.8 Flash serves as a guiding agent in the loop: as Genie 4 generates a world in real-time, Gemini continuously observes the scene.' Credit goes to Raia Hadsell, Diego Rivas and Hang Qi.

Access is gated. Genie 4 is a research preview for trusted testers, rolling out first to Google AI Ultra subscribers through Project Genie, with early access partners of the Gemini API and Google AI Studio also included. Everything it emits carries SynthID, DeepMind's imperceptible watermark for AI-generated content.

The post is candid about the limits: 'visual details and object permanence can still shift or drift over time—especially as a simulation runs longer, after multiple complex prompt edits, or when navigating throughout the scene.' No resolution, frame-rate, inference cost or maximum session length appears in the write-up. It drops into a steady run of Google releases we have logged on our Google AI feed.