Google Genie 4 world model adds sound, live scene editing and characters that talk back

What Genie 4 does

Genie 4 creates a simulated world from a prompt and lets the user explore it and interact with it as it's generated. It follows Genie 3, which DeepMind announced in August 2025.

Genie 4 adds sound. DeepMind calls it its first real-time world model to generate audio natively, so ambient sound, music and speech are created along with the visuals. Characters in a scene can hold conversations and react to what the user says or does.

Users can also edit a scene while it runs. A plain-language request can alter the weather, move the camera or add people and objects. DeepMind says the model improves photorealism, movement and physics such as collisions and aerodynamics, and makes human characters more realistic.

How Gemini fits in

Genie 4 leans on Google's Gemini models. According to DeepMind, Gemini Omni keeps simulations grounded in how the real world looks and behaves, while other Gemini models handle the reasoning that keeps a scene consistent. Gemini also adapts users' prompts to the scene in progress.

DeepMind also describes Gemini 3.8 Flash acting as a guiding agent. It breaks a goal into steps, checks the generated frames for progress and changes its plan when the world shifts. One demo shows Gemini 4 Argon and Genie 4 simulating Apptronik's Apollo humanoid robot playing checkers.

Who can use it and when

Genie 4 is rolling out to a limited group of trusted testers first. Google says it is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners using the Gemini API and Google AI Studio. DeepMind has a waitlist for Genie 4 in Project Genie.

Google hasn't given a release date, a price or a list of countries. Testers have used the preview to plan film shots, turn a 2D photo into a simulated scene and prototype game ideas, according to DeepMind.

Known limits

DeepMind says visual details and objects can drift during long sessions, after many edits or as the user moves around. It lists complex physical interactions, accurate real-time audio, readable text inside scenes and long-term consistency as areas it is still working on.

All video and audio that Genie 4 generates carry SynthID, Google's watermark for identifying AI-generated content.

What it means for you

Most people can't try Genie 4 yet. The first wider release is for Google AI Ultra subscribers and early access developer partners, and Google hasn't said when it might reach other plans.

Google is pitching Genie 4 at filmmakers, game makers and other people who build simulations. A scene that responds to typed direction and comes with its own sound could make it faster to mock up an idea. Google also describes world models as safe training environments for robots and AI agents.

Sources

• Google DeepMind: Genie 4, a new generation of world simulation
• Google: We're announcing Genie 4, our new world model for real-time interactive simulation
• TechCrunch: Google's Genie 4 world model adds audio, more realistic worlds and real-time steering