DeepMind Genie 4 Adds Real-Time Audio

Google DeepMind on Oct. 27 unveiled Genie 4, a real-time world model that generates sound alongside the interactive scenes it builds, with Gemini 3.8 Flash acting as a guiding agent that keeps each simulation on track. The model is a research preview for trusted testers, and Google said it is "coming soon" to AI Ultra subscribers in Project Genie without giving a date.

Key takeaways

• Genie 4 is DeepMind's first real-time world model to generate audio natively, including ambient sound and character dialogue that adapts on the fly.
• Gemini 3.8 Flash watches the scene, breaks goals into steps, checks incoming frames and revises its plan when a user intervenes.
• Access is limited to trusted testers for now. AI Ultra subscribers and early-access Gemini API and AI Studio partners are next, with no date announced.

What is Genie 4?

A world model generates an explorable environment frame by frame and responds to what the user does inside it, rather than producing a fixed video clip. In DeepMind's demos, users navigate with keyboard controls and type prompts to change the scene.

According to the DeepMind announcement, written by Raia Hadsell, Diego Rivas and Hang Qi, users can change the weather, move the camera or add a person or object mid-simulation. Gemini rewrites each prompt to fit the surrounding scene, and characters move and react to their surroundings.

Genie 4 audio: sound and dialogue in real time

"Genie 4 is our first real-time world model to generate audio natively," DeepMind wrote. Each world gets its own ambient sound, and characters hold dialogue that changes as the scene does. The company also says the model simulates aerodynamics and collisions, though it gave no accuracy figures.

How Gemini steers the simulation

DeepMind describes Genie 4 as part of a stack of Gemini models. Gemini 3.8 Flash runs as a guiding agent during generation: it observes the scene, splits a goal into steps, reviews new frames to confirm each step worked and adjusts when the world shifts. Between user inputs, Gemini keeps the scene coherent.

The post credits Gemini Omni with grounding the look and behavior of simulations in the real world. In one robotics demo, a model DeepMind calls Gemini 4 Argon breaks a task into steps that are fed to Genie 4 as interleaved text captions, showing an Apptronik Apollo humanoid playing checkers.

Context: from Genie 3 to Project Genie

Genie 4 follows Genie 3, announced Aug. 5, 2025, which ran at 720p and 24 frames per second and stayed consistent for a few minutes. Google opened Project Genie, a Labs prototype built on that work, to U.S. AI Ultra subscribers aged 18 and older on Jan. 29, 2026, with sessions capped at 60 seconds, The Decoder reported.

Who is testing it, and the known limits

Trusted testers come from creative industries, journalism, education, gaming and engineering. Reported uses include previsualization "takes," turning a single 2D photograph into a simulated environment, visualizing flash floods or wildfires, and rapid game prototyping.

DeepMind acknowledged that "visual details and object permanence can still shift or drift over time," especially in longer sessions or after many prompt edits. It lists complex physical interactions, real-time audio accuracy, legible text and long-term consistency as open research problems. All Genie 4 video and audio will carry SynthID, Google's invisible watermark.

What's next

Google plans to roll Genie 4 out to AI Ultra subscribers in Project Genie and to early-access partners in the Gemini API and Google AI Studio, with a waitlist now open. DeepMind says the longer-term goal is "hybrid world models," in which Gemini plans the narrative and Genie renders the scene. API pricing and availability outside the preview have not been announced.

FAQ

Can I use Genie 4 today?

Not unless you are a trusted tester. Google says it is coming soon to AI Ultra subscribers in Project Genie and to early-access partners in the Gemini API and Google AI Studio, but it has not given a date.

What is new compared with Genie 3?

Genie 4 generates audio natively and is steered in real time by Gemini 3.8 Flash. DeepMind has not released comparable resolution or frame-rate figures.

How will Genie 4 content be labeled?

All video and audio generated by Genie 4 will include SynthID watermarking, according to DeepMind.

References

• Genie 4: A new generation of world simulation — Google DeepMind (Oct. 27, 2026)
• Genie 3: A new frontier for world models — Google DeepMind (Aug. 5, 2025)
• Google DeepMind opens Project Genie to US subscribers for real-time AI world generation — The Decoder (Jan. 29, 2026)