Google DeepMind unveils Genie 4, its first world model with native audio

Google DeepMind released Genie 4 in research preview on October 27, 2026, its first real time world model to generate audio natively. Genie 4 builds explorable, photorealistic worlds from text or image prompts and generates the soundscape with them: music, ambient sound, and character dialogue. Users navigate with keyboard controls and can steer a running simulation with typed prompts, changing the weather, the camera, or adding people and objects, and scenes stay coherent for up to about two minutes, roughly double the Genie 3 limit. A Gemini model runs inside the loop as a director, planning steps and checking frames, and output carries SynthID watermarking.

Access starts with trusted testers; Google says Google AI Ultra subscribers in Project Genie and Gemini API and AI Studio early access partners follow. Resolution, frame rate, and pricing were not published.

Sources: TechCrunch; MarkTechPost