Google Launches Genie 4 World Model With Native Audio and Real-Time Steering for Trusted Testers

Google (GOOGL) introduced Genie 4 on Tuesday, a new world model from its DeepMind unit that generates interactive simulated worlds in real time, now with audio created natively alongside the visuals. Google calls it its first real-time world model to generate audio natively. The model is in a research preview for a limited group of trusted testers, with no general release date.

What Genie 4 does

Genie 4 builds a simulated world that a person can move through and change as it runs. Users navigate and control the camera with standard keyboard controls and steer the scene with natural-language prompts, such as changing the weather, moving the camera or adding characters. Google says Gemini adapts those prompts to the scene and watches it between inputs to keep it coherent. The company said it simulates physics such as aerodynamics and collisions, and that characters move naturally, respond to their surroundings and can hold conversations that react to user input or events.

The audio is new. Music, ambient sound and interactive dialogue are produced in real time rather than added afterward. DeepMind said Gemini Omni anchors the simulations in real-world appearance and behavior, while other Gemini models handle the reasoning needed for continuity. Gemini 3.8 Flash acts as a guiding agent that breaks goals into steps and checks incoming frames as it goes.

How to get it

Where Google sees it used

Google pointed to training environments for robots and agents, including a demo of a humanoid robot from Apptronik playing checkers in simulation, as well as creative uses such as previsualization for filmmakers, game and interactive prototyping, and turning a single archival photo into a simulated environment. DeepMind also framed world simulation as a step toward more general AI.

Limits and safeguards

Google describes Genie 4 as early and experimental. Visual details and object permanence can drift, particularly in longer sessions, after several prompt edits or while moving through a scene, and the company lists complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency as open research areas. It did not publish resolution, frame rate or how long a simulation stays consistent. For comparison, its earlier Genie 3 model, announced in August 2025, ran at 720p and 24 frames per second and held a world together for a few minutes. Google said all generated video and audio carries a SynthID watermark and that it ran internal safety evaluations before the preview.

What it means for investors

World models are an early-stage area for Alphabet, with no announced pricing or revenue tied to Genie 4. The launch matters mainly as a sign of how DeepMind is extending its Gemini models into simulation for robotics, media and gaming. The things to watch are when Genie reaches paying AI Ultra subscribers and developers, and whether Google publishes performance details.

Sources: Google DeepMind, Genie 4: a new generation of world simulation (Oct. 27). This is informational, not a recommendation to buy or sell.