Google DeepMind releases Genie 4, a world model with native audio and real-time steering

The short version

Google DeepMind released Genie 4, a world model it says can generate real-time interactive worlds with native audio, physics, and human movement, and steer scenes through natural language. The model begins as a research preview for trusted testers.

What changed

Genie 4 is a new world-simulation model. Google says it generates visual worlds with ambient and dialogue audio while Gemini models help with real-time steering and continuity.

What it means for you

This is a new model category for interactive world generation. Creative, game, education, robotics, and agent teams can assess prototyping value, while accounting for the preview's reliability, availability, and safety limits.

Google DeepMind has released Genie 4, a new world model. The company says it can generate explorable, steerable interactive worlds in real time, while producing ambient sound, music, and interactive dialogue alongside the visual scene. Users can navigate and control a camera with standard keyboard controls, then alter weather, camera position, people, or objects with natural-language instructions as a scene evolves; Google says Gemini models adapt those instructions to the surrounding scene.

This is a new model-category release rather than a routine feature update to an existing tool. It creates an interactive-scene generation path for previsualization, game prototyping, education, robotics training environments, and agent evaluation. Google says Genie 4 combines physics simulation, natural human motion, native audio, and real-time steering, and that it is collecting feedback with trusted industry testers.

Availability is still limited. The announcement says Genie 4 starts as a research preview for trusted testers, with planned initial access for Google AI Ultra subscribers in Project Genie and early-access partners for the Gemini API and Google AI Studio. Teams considering it should first confirm access eligibility, data-handling terms, and workflow fit instead of treating the announcement as a generally available service.

Google describes the model as early and evolving research. Visual details and object permanence can shift or drift during longer sessions, after several complex prompt edits, or while navigating a simulation. Complex physical interaction, real-time audio accuracy, readable text, spatial accuracy, and long-term consistency remain active work. Google says every video and audio output generated by Genie 4 will include a SynthID watermark. Validate output consistency and safety procedures in a controlled, reversible prototype, and do not treat a simulated result as a guarantee of real-world behavior.

Related directory entry

Fact check

VerifiedGoogle DeepMind announced Genie 4 as a new world model for real-time interactive world simulation.Evidence

VerifiedGoogle says Genie 4 generates audio natively and adds physics simulation, human movement, and real-time natural-language scene steering.Evidence

VerifiedGoogle says Genie 4 begins as a research preview for trusted testers, before planned availability to Google AI Ultra subscribers in Project Genie and early-access partners for the Gemini API and Google AI Studio.Evidence

VerifiedGoogle says every Genie 4-generated video and audio output carries a SynthID watermark and that visual details and object permanence can drift in longer sessions or after complex edits.Evidence