Genie 4: Google's world model gains native audio and live steering
Its most photorealistic sim holds a scene for two minutes. Trusted testers only today; Gemini API and AI Studio access is coming soon. Output is SynthID-marked.
Copy markdown
Native audio and two-minute worlds
Genie 4 generates music, ambient sound, and interactive dialogue alongside the visuals — the first real-time world model to do audio natively — and holds a scene coherent for up to two minutes, double Genie 3's window.
Steer a running sim in plain language
Type mid-simulation to change the weather, move the camera, or drop in a person or object. Gemini 3.8 Flash acts as a 'director,' breaking goals into steps and checking frames while Genie renders the world.
Waitlist now; Gemini API access is next
Today it's trusted testers only, behind a waitlist. Google AI Ultra subscribers get it through Project Genie, and Gemini API plus AI Studio early-access partners are up next — both 'coming soon.' No public API or pricing yet.
Every frame ships SynthID-watermarked
All Genie 4 video and audio carry SynthID watermarks, so provenance marking is baked into anything you build on it. Known rough edges: in-world text and audio accuracy still drift the longer a session runs.