Google DeepMind announced Genie 4 on October 27, 2026, and it is the first world model that lets you walk through a computer generated world, change it with plain text while it runs, and hear it talk back. The demo clips show photorealistic scenes with real physics, ambient sound, and characters who hold conversations with you in real time.
This post explains what Google Genie 4 actually is, the four capabilities DeepMind highlighted, who gets access first, and how to join the waitlist. We also cover what early testers and researchers are saying about it.
What Is Google Genie 4?
Genie 4 is a world model, meaning an AI system that learns to simulate environments rather than just generate static images or clips. You type what you want to see, and Genie 4 builds a playable, explorable world that keeps going, with objects and people that follow consistent rules.
DeepMind says Genie 4 is natively multimodal. Music, ambient noises, and interactive dialogue are generated together with the world instead of being added later. The model builds on the Gemini family of models: Gemini Omni anchors the simulations in how the real world looks and behaves, while other Gemini models handle the reasoning that keeps a simulation coherent over time.
Announced by Raia Hadsell, Diego Rivas and Hang Qi on the DeepMind blog, Genie 4 arrives about 14 months after Genie 3, which DeepMind called a new frontier for world models in August 2025.
The Four Capabilities DeepMind Highlighted
1. Shaping the world in real time
With Genie 4 you can modify a simulation while it is running. Type that it started raining, move the camera, or introduce a new person or object, and Gemini adapts your prompt to match the surrounding scene. Between your inputs, Gemini also monitors the scene and keeps the world progressing coherently. Navigation uses standard keyboard and mouse controls, and text prompts steer what happens next.
2. Physics and human action
DeepMind says Genie 4 simulates real-world physics scenarios such as aerodynamics and collisions. The bigger leap is how it handles people. Characters move naturally, respond to their environment, and can be steered interactively in real time.
3. Native audio and interactive dialogue
Genie 4 is DeepMind’s first real-time world model to generate audio natively. Every world carries the ambient sound of its setting, from wind in the trees to waves at a beach. Characters also speak. Their dialogue adapts on the fly as you interact with them or as events unfold.
4. Complex events through Gemini reasoning
For multi-step scenes, Gemini 3.8 Flash works as a guiding agent in the loop. As Genie 4 generates the world in real time, Gemini watches the scene, breaks goals into steps, and reviews incoming frames to check that each step succeeded. If the world shifts or you intervene, Gemini adapts its plan. DeepMind says this is what gives simulations reliable causality, procedural behaviour, and multi-step reasoning in real time.
How Genie 4 Is Different From Text-to-Video Tools
Text-to-video models generate a finished clip from a prompt. Genie 4 does something structurally different: it generates an environment you can explore while it is being created. The camera is yours to move, the world keeps evolving when you stop typing, and characters continue to act on their own.
Three things separate it from video generation:
- Persistence: the scene stays consistent as you move through it, because Gemini models track the simulation’s state and continuity
- Live steering: changes are applied to a running world instead of forcing a full regeneration
- Native sound: audio and dialogue are produced with the world, not mixed in afterwards
DeepMind frames this as a step towards AGI, since understanding and simulating the world is part of training systems that can act inside it. The practical near-term value is in training environments for robots and agents, plus interactive experiences that did not exist before.
Who Can Use Genie 4 Right Now?
Genie 4 is in research preview for trusted testers. DeepMind says it will then release it to consumers and developers in this order:
- Google AI Ultra subscribers, inside a product experience called Project Genie
- Early access partners of the Gemini API and Google AI Studio
- A public waitlist is open now through a Google Forms link on the DeepMind blog
Logan Kilpatrick of Google also confirmed on X that Genie 4 is coming to the API “for the first time”, which matters for developers who want to build on world models instead of just watching demos.
What Early Testers Are Using It For
DeepMind shared a first look at how trusted testers are working with Genie 4:
- Film and creative pre-visualization: filmmakers step into a live simulation of a scene and run multiple takes in real time, adjusting camera angles, lighting, dialogue and action on the fly
- Robot training: DeepMind is pairing Genie 4 with Gemini 4 Argon and Gemini Robotics. In one example, Gemini 4 Argon breaks a task into steps that Genie 4 renders, producing a simulation of Apptronik’s Apollo humanoid playing checkers
- Education: interactive reconstructions, such as historical scenes you can walk through
- Enterprise simulation: partners are using it for rapid scenario generation to test processes and environments before building them in the real world
Why It Made News Beyond AI Circles
Bloomberg reported that the upgrade “rattled video game stocks”, because a model that generates playable worlds on demand could change how game content gets made. Ars Technica’s Benj Edwards posted on X that “we’re getting closer and closer to the ‘neural rendering’ era of gaming”.
Ethan Mollick of the Wharton School, who tested Genie 4, posted on X: “Genie 4 is quite an insane experience, generating game-like enviornments as you go. It isn’t flawless but I can type ‘Neogothic brutalist city inhabited by robots and horses’ & it creates one in a minute.”
Sundar Pichai highlighted four things in his own post on X: steering dynamic environments in real time, native audio matched to the world, photorealistic scenes with real-world physics, and more coherent multi-step events through Gemini reasoning.
Tips If You Want to Try Genie 4
- Join the waitlist now from the DeepMind blog. Access is rolling out in phases, and Ultra subscribers are next in line
- Watch the Project Genie space in your Google AI Ultra account for the invite when it opens
- Study the prompt style in DeepMind’s demo clips. They describe the scene like a film brief: subject, setting, camera behaviour, lighting, and audio, all in one prompt
- Use short steering prompts while the world runs (“change the weather to snow”, “add a friendly otter character”) instead of rebuilding the world from scratch
- Developers should watch the Gemini API changelog, since Genie 4 access through the API is planned for early access partners first
FAQ
What is Google Genie 4?
Genie 4 is Google DeepMind’s newest world model, announced on October 27, 2026. It generates interactive, playable environments in real time with photorealistic visuals, physics, native audio, and characters you can talk to and steer with text prompts.
How can I try Genie 4?
Genie 4 is in research preview for trusted testers. Google AI Ultra subscribers get access next through Project Genie, and developers can get early access through the Gemini API and Google AI Studio. A public waitlist is open on the DeepMind blog.
Does Genie 4 generate sound and voices?
Yes. Genie 4 is DeepMind’s first real-time world model to generate audio natively, including ambient sounds, music, and interactive dialogue. Characters can hold conversations with you as the world runs.
What is new in Genie 4 compared to Genie 3?
Genie 4 adds native audio, better physics and photorealism, real-time text steering of live scenes, and multi-step event reasoning powered by Gemini 3.8 Flash. Genie 3, released in August 2025, generated worlds primarily as visual environments with basic controls.
Is Genie 4 available for free?
There is no public pricing yet. The first consumer access is planned for Google AI Ultra subscribers, which is Google’s top paid AI tier. Developers should expect API pricing when Gemini API access opens.
