Google has introduced Genie 4, a world model for interactive simulations that produces native audio and lets users direct scenes through natural language as they run. Users can alter weather, add characters or reposition the camera while a simulation unfolds, according to Google.
Google has begun giving a limited group of trusted testers access to Genie 4. The company says Google AI Ultra subscribers will get access soon through Project Genie, while early access partners will receive it through the Gemini API and AI Studio.
Read more: Google Launches Public Testing for AI World Builder Project Genie
The model generates sound during the simulation, from environmental noise and music to dialogue that responds to interactions. Google says Gemini Omni helps ground the simulated environments in how the real world looks and behaves, with other models in the Gemini family supplying reasoning to keep events consistent over time.
The ability to introduce characters during a running simulation marks a change from the earlier Project Genie prototype, which lacked the demonstrated capability to insert elements into an ongoing world. That version used Genie 3 and offered 60-second sessions, with video output at 720p and up to 24 frames per second.
Google sees world simulation as a step toward AGI and says it could provide safe environments in which robots and agents can train. The company also points to potential creative uses in gaming and filmmaking, alongside applications in journalism and manufacturing.
