Skip to main content
Back to timeline
Google DeepMindSource publication:

DeepMind's Genie 4 world model generates ambient sound and character dialogue alongside real-time video

Synopsis

Google DeepMind released Genie 4, a world model that generates visuals, ambient sound and character dialogue together in real-time interactive simulation, using Gemini models to keep a scene coherent. Genie 3 had already pushed real-time world models forward in August 2025, but audio was not natively generated. Genie 4 opens first as a research preview to trusted testers, and DeepMind says visual details and object permanence can still drift over longer runs or after repeated complex prompt edits.

Source-provided article image: Genie 4: A new generation of world simulation — Google DeepMind
Google DeepMind

Interpretation

Genie 4 extends real-time world simulation from images to sound and speech: every world carries the ambient sound of its own setting, and characters can converse and react to user inputs or events in the world in real time. DeepMind had released Genie 3 in August 2025 as a new frontier for world models; Genie 4 is the company's first real-time world model to generate audio natively. DeepMind demonstrates the behavior with recordings of real-time interactions in which users navigate and control the camera with standard keyboard controls and steer the scene with text prompts, and all accompanying sounds in the demos are produced directly by Genie 4.

Coherence is handled behind the scenes by Gemini models: Gemini Omni anchors the simulation in real-world appearance and behavior, while other Gemini models supply the reasoning needed to maintain a simulation's coherence and continuity. Gemini 3.8 Flash acts as a guiding agent in the loop, observing the scene as Genie 4 generates it, breaking goals into a sequence of steps, and reviewing incoming frames to track progress and verify that each step succeeded. Gemini adapts its plan in the moment as the world shifts or the user intervenes; DeepMind also experimented with Gemini 4 Argon and Genie 4 together, breaking a user-instructed task into steps interleaved as text captions to simulate Apptronik's Apollo humanoid playing checkers.

Perspective

Genie 4 is currently a research preview for trusted testers, with later access planned for Google AI Ultra subscribers in Project Genie and early access partners in the Gemini API and Google AI Studio. DeepMind describes tester scenarios spanning creative-industry pre-visualization, turning a single 2D archival photograph into a grounded simulation of that environment in seconds, game prototyping and cross-industry simulation scenario generation.

DeepMind states that Genie 4 remains an early, experimental model: it can maintain coherent environments over short sessions, but visual details and object permanence can shift or drift over time, especially as a simulation runs longer, after multiple complex prompt edits, or when navigating through the scene. Complex physical interactions, real-time audio accuracy, rendering legible text, spatial accuracy and long-term consistency are listed as active research areas.