Google DeepMind Launches Genie 4, an Interactive World Model with Live Audio

Google DeepMind unveiled Genie 4 on Tuesday, making it available immediately to trusted testers, with a broader rollout to Google AI Ultra subscribers via Project Genie and early access partners of the Gemini API and Google AI Studio expected in the coming weeks. Genie 4 is a world model — it generates interactive virtual environments from text or image prompts — and represents a major upgrade over its predecessor in three key areas. First, it generates native, synchronised audio: music, ambient noises, and reactive dialogue from characters all appear alongside the simulated world in real time. Second, it doubles its predecessor's scene-consistency limit, maintaining coherent physics and object placement for up to two minutes. Third, users can steer simulations mid-run by typing prompts to change weather, camera angles, or add objects, making the experience genuinely interactive rather than a one-shot generation. Genie 4 is tightly integrated with Gemini, Google's AI model, which acts as the director that plans steps and checks frames while Genie creates the visuals. In a related robotics experiment, a Gemini 4 Argon model broke down tasks into steps that Genie 4 simulated in real time, pointing toward testing robot scenarios virtually before real-world deployment. Potential applications cited by the team include game prototyping, robotics simulation, film storyboarding, and interactive learning. Current limitations include occasional drift in fine details during longer simulations and challenges ensuring audio accuracy and in-world text legibility.