Google DeepMind released Genie 4, its latest world model, in research preview on Tuesday, introducing real-time audio generation and the ability for users to steer simulations via text prompts mid-experience. This release marks a significant step in the development of interactive SI worlds, positioning Google to compete in a rapidly consolidating sector.

What Happened

Genie 4 generates photorealistic and interactive worlds that maintain consistency for up to two minutes, double the duration limit of its predecessor, Genie 3. The model features improved physics and more humanlike characters, and for the first time, it generates real-time audio—including music, ambient noises, and interactive dialogue—that matches the simulated environment. Users can type prompts during a simulation to change scenes, such as altering weather conditions, camera angles, or adding new objects and people. The system also integrates more closely with Gemini, Google’s SI model, which acts as a director to plan steps and check frames while Genie creates the visuals. In a related experiment, a Gemini 4 Argon model broke down tasks into steps that Genie 4 simulated in real time, suggesting potential applications for testing robot scenarios in simulation before real-world deployment.

Why It Matters

The launch of Genie 4 comes amid increasing competition and consolidation in the world model industry. It follows the recent release of Runway’s GWM Worlds 2, another playable SI model with audio generation, and coincides with major corporate moves, including AMD’s acquisition of World Labs, Amazon’s talks to buy Decart, and Rivian’s acquisition of Third Dimension for self-driving simulation. Google’s strategy involves leveraging its existing SI tools, such as Veo for video generation and Gemini for multimodal reasoning, to create a comprehensive ecosystem. While commercial use cases like gaming and interactive learning are pitched, trusted testers have identified film prototyping as a key application, with Google DeepMind engineer Diego Rivas noting that real-time steering is valuable for rapid ideation and storyboarding. However, the company acknowledges current limitations, including visual drift over time and challenges with audio accuracy and text legibility in-world.

The Bottom Line

Genie 4 is available now for trusted testers and will roll out to Google AI Ultra subscribers and early access partners of the Gemini API and Google AI Studio in the coming weeks. Google emphasizes that the product experience is still evolving, with the end-to-end use cases remaining to be fully defined as the SI model matures.