Google Open-Sources Genie4 World Model API with Real-Time Interaction and Native Audio Generation
1 day ago / Read about 0 minute
Author:小编   

Google has officially opened its latest world model, Genie4, to paid developers. This marks Google DeepMind's first real-time world model that third-party developers can call directly, no longer relying on consumer subscriptions or demo environments. Built on Gemini Omni and capabilities from the Gemini series, the model can generate continuous, high-definition interactive worlds at 720p and 24fps, with synchronized native stereo audio output. Users can control movement in real time and dynamically adjust scene elements, while the model maintains coherent world states; generated content is embedded with SynthID watermarks. On the security side, it uses single-session temporary tokens and transmits streaming media via WebRTC and WebSocket, with a single session capped at 120 seconds. The service is available in data centers across multiple regions, and corresponding API billing standards have been published. Genie4 pushes world simulation capabilities into the third-party ecosystem and can serve a wide range of scenarios.