Genie 4: DeepMind's world model now generates sound too
Genie 4 from Google DeepMind generates interactive worlds in real time, complete with music, ambient sound and character dialogue. Here's what else it promises.
This article in other languages:العربيةDeutschEspañolFrançaisItaliano日本語Polski简体中文
In this article
On October 27, Google DeepMind announced Genie 4, the new generation of its world model: an AI that generates an interactive world in real time that you can move around in. For now it’s a research preview for selected testers, with Google AI Ultra subscribers in Project Genie and Gemini API and Google AI Studio partners next in line.
What’s new in Genie 4?
The headline feature is sound. DeepMind calls it “our first real-time world model to generate audio natively”. Alongside the picture, it generates:
- ambient sound,
- music,
- character dialogue that adapts to what’s happening in the moment.
The second change is steering the world with plain text while the simulation runs. We can change the weather, move the camera, or add a person or an object, all without starting over. Gemini models map the prompt onto the scene and keep events going between prompts so the world stays coherent.
The third is physics and people. The announcement mentions simulated aerodynamics and collisions, and characters that move on their own and react to their surroundings.
How does Genie 4 work with Gemini?
This is where it gets interesting for agent fans. Genie 4 builds on the Gemini family: Gemini Omni makes sure the world looks and behaves like a real one, while other Gemini models handle reasoning and continuity.
In the demo, Gemini 3.8 Flash acts as an agent inside the generated world: it observes the scene, breaks a goal into steps, checks on the following frames whether each step worked, and fixes its plan when the world changes or the user steps in. DeepMind also showed an experiment with Gemini 4 Argon and Gemini Robotics in which Genie 4 simulates Apptronik’s Apollo humanoid playing checkers (no physical robot involved).
It’s the same direction as the Gemini agents that beat Minecraft, except here the world doesn’t come from a game but from a model.
How did Genie 3 compare?
For reference, Genie 3 from August 2025 generated worlds at 720p and 24 frames per second that stayed consistent for a few minutes, with visual memory reaching back up to a minute. For Genie 4, DeepMind hasn’t shared such specs (resolution, frame rate, session length) yet.
Genie 3 later landed in Project Genie for AI Ultra subscribers and shook up the stock market. On January 30, 2026, Sherwood News reported that Unity shares were down more than 25% and Roblox about 9%. Hence the question that’s back with Genie 4: will world models threaten game engines? For now it’s more of a curiosity than a ready tool for making games ;)
How can you try Genie 4?
Right now you can’t try it yourself unless you’re one of the invited testers (DeepMind lists the creative industries, journalism, education, gaming and engineering). After that, the plan looks like this:
- Project Genie for Google AI Ultra subscribers, “soon”.
- Gemini API and Google AI Studio for early-access partners.
- A waitlist for Project Genie, linked in the announcement.
DeepMind hasn’t given dates or pricing for API access.
What should we watch out for?
First, the limitations DeepMind lists itself. In longer sessions and after many edits, image details can shift and objects don’t always stay where they were. Complex physical interactions, real-time audio accuracy, legible text and long-term consistency are still open problems.
Second, the lack of numbers. We have a description of the capabilities and demos, but no technical specs, so it’s hard to judge how big the jump from Genie 3 is.
Third, content labeling. According to DeepMind, every video and audio output from Genie 4 will carry a SynthID watermark.
FAQ
What is Genie 4?
It’s a world model from Google DeepMind that generates an interactive environment in real time, with picture, sound and character dialogue. You can change the world with text prompts while the simulation runs.
When will Genie 4 be available?
Since October 27, 2026, selected testers have been using it. According to DeepMind, it’s coming “soon” to Google AI Ultra subscribers in Project Genie and to Gemini API and Google AI Studio partners. No exact date has been given.
Is Genie 4 free?
Nothing like that has been announced. Consumer access is set to go through the paid Google AI Ultra subscription, and API pricing hasn’t been shared.
How is Genie 4 different from Genie 3?
According to DeepMind, Genie 4 generates sound natively, simulates physics and people better, is more photorealistic, and lets you steer more complex, multi-step events in real time.