Google Genie 4 Adds Real-Time Audio, Live World Editing and Longer Simulations

Summary

Google DeepMind released Genie 4 on October 27, 2026, as a research preview for trusted testers. Its most photorealistic world model yet generates interactive scenes with synchronized music, ambient sound and dialogue. Users can change weather, camera angles, people or objects through prompts during simulations. Scenes remain consistent for up to two minutes, twice Genie 3’s limit, with improved physics and more humanlike characters.

Gemini directs the process by planning steps and checking frames while Genie 4 renders visuals. In a Gemini Robotics experiment, Gemini 4 Argon divided tasks into steps that Genie 4 simulated live, suggesting robots could be tested virtually before physical trials. Research director Tejas Kulkarni expects hybrid systems combining Gemini’s code and text with Genie simulations. Proposed uses include gaming, robotics, interactive learning, conversations with figures inside historical paintings and film prototyping. Engineer Diego Rivas said real-time control supports rapid ideation and storyboarding. Google’s Veo already generates high-definition clips with synchronized audio, while eventual conversion of prototypes into assets through multimodal Gemini Omni remains distant.

Genie 4 still suffers object and detail drift, inaccurate audio and illegible in-world text, and Google has not settled its final product experience. Access will expand in coming weeks through Project Genie for Google AI Ultra subscribers and to early Gemini API and Google AI Studio partners. Competition includes Runway’s playable, audio-generating GWM Worlds 2, released months earlier. Consolidation is accelerating as AMD acquires Fei-Fei Li’s World Labs, Amazon discusses buying Decart, and Rivian acquires Third Dimension for self-driving simulation.