Google DeepMind has introduced Genie 4, an experimental artificial intelligence model that builds interactive virtual worlds from text descriptions. The updated system allows users to alter environments in real time using text commands while generating native ambient audio and spoken character dialogue. Access is currently limited to trusted research testers, with plans to expand availability to Google AI Ultra subscribers and API partners.
Users can alter active simulations on the fly with basic text prompts. The system accepts commands to change weather conditions, reposition camera angles, or insert new characters and props into the ongoing scene. Google refined the underlying engine to deliver more accurate collisions, aerodynamic forces, and motion tracking. Onscreen characters move more naturally, react directly to environmental changes, and adapt their actions as the simulation unfolds.
Genie 4 is the first release in the family to generate audio natively alongside visual environments. The model synthesizes sound effects like wind and running water directly into the space. Characters can speak, answer user questions, and react verbally to events occurring inside the scene without secondary audio tools.
Google is distributing Genie 4 as a closed research preview for selected testers. The rollout will next expand to Google AI Ultra subscribers through Project Genie. Developers and enterprise partners will later gain access to the system through the Gemini API and Google AI Studio.
