Google Genie 4: world simulation model

Google launches Genie 4, a world simulation model with realism and real-time interaction. The new version integrates Gemini Omni and Gemini 3.8 Flash to maintain consistency and physics. Availability is limited to testers and Project Genie subscribers.

Genie 4 enables simulations for robotics, entertainment and education. The version can maintain a consistent scene for up to two minutes, doubling the limit of Genie 3. The version is part of efforts to create hybrid world models, combining logic and sensory simulation. It is aimed at developers and Google AI Ultra users, with applications in robotics, entertainment and education. It is offered to limited testers and will be made available to Project Genie subscribers and to Gemini API and AI Studio partners.

Integration with Gemini ensures realism and natural reactions

Integration with natural language lets users create worlds with simple prompts, without needing technical knowledge. Native audio generation makes dialogue and sounds more engaging. Genie 4 gains realism and real-time interactivity for developers and researchers. The combination of these features allows users to explore virtual worlds with greater depth and immersion.

Google DeepMind reveals that Genie 4 simulates characters that move naturally and react to their environment.

According to Raia Hadsell, Diego Rivas and Hang Qi, "the biggest advance is in how Genie 4 creates characters that move organically, respond to the environment and can be guided in real time" (loosely translated).

The new version also generates native audio, such as the sound of crashing waves.

Genie 4 extends consistent scenes and includes natural language for more realistic simulations. Developers can test complex scenarios without specialized hardware. AI-generated worlds can be edited in real time, with improved dialogue and physics. This reduces development time and increases creative flexibility.