Trend · AI models
Genie 4: Google's AI that builds a world you can walk into
Updated Oct 28, 2026 · By Mishal Brands Editorial
Checked Oct 28, 2026. Every claim below comes from Google DeepMind's own announcement on Oct 27, 2026, which is linked at the bottom. We have not used Genie 4.
The short answer
Genie 4 is Google DeepMind's new world model: an AI that builds a place you can move around in, in real time, with sound. DeepMind announced it on Oct 27, 2026. You steer with ordinary keyboard controls and with text prompts, and the world keeps going between your inputs. It is the first real-time world model from DeepMind that generates audio natively, including characters you can talk with. Right now it is a research preview for trusted testers. DeepMind says it is coming soon to Google AI Ultra subscribers in Project Genie and to early-access partners in the Gemini API and Google AI Studio, and the waitlist is open.
What it actually does
DeepMind describes four capabilities, and each one is easier to picture than it sounds.
- You shape the world while it runs. With standard keyboard controls you move and turn the camera. With text you change the weather, move the camera, or bring in a new person or object. DeepMind says Gemini adapts your prompt to fit the scene around it.
- It simulates physics and people. DeepMind says Genie 4 handles aerodynamics and collisions, and that the biggest leap is in people: characters move naturally, respond to their surroundings, and can be steered in real time.
- It makes its own sound. This is DeepMind's first real-time world model to generate audio natively. Every world carries the ambient sound of its setting, from wind in trees to waves on a beach. Characters also hold interactive dialogue that adapts on the fly.
- Gemini keeps the story straight. Gemini 3.8 Flash works inside the loop as a guiding agent: it watches the scene, breaks a goal into steps, checks incoming frames to see whether each step worked, and replans when you intervene or the world changes.
How it is built
Genie 4 borrows from Google's Gemini models rather than starting from nothing. DeepMind says Gemini Omni anchors a simulation in real-world appearance and behavior, while other Gemini models supply the reasoning that keeps a world coherent as it runs. DeepMind also showed Gemini 4 Argon breaking a task into steps for Genie 4 to render, in an experiment with Gemini Robotics where a simulated Apptronik Apollo humanoid plays checkers.
Who gets it, and when
| Where | Status on Oct 28, 2026 |
|---|---|
| Trusted testers | Research preview is already rolling out. DeepMind says it is gathering feedback before a wider release. |
| Google AI Ultra subscribers, in Project Genie | Coming soon, according to DeepMind. The waitlist is open, and the sign-up link is in DeepMind's announcement. |
| Gemini API and Google AI Studio | Coming soon for early-access partners. |
| Everyone else | No date. DeepMind has not announced general availability or pricing for Genie 4. |
What testers are doing with it
DeepMind has been running Genie 4 with testers in creative industries, journalism, education, gaming and engineering. It describes four patterns:
- Draft mode for film and video. A director can step into a live simulation of a scene and run several takes in a row, changing camera angles, lighting, dialogue and action as it goes.
- A photograph you can walk into. Photojournalists are testing turning a single 2D archival photo into a grounded simulation of that place in seconds.
- Game prototypes without an engine. Developers are going from an idea to an explorable prototype in seconds, to test a visual style or a mechanic before writing any code.
- Scenario generation. DeepMind says enterprise partners are using it to spin up situations quickly, including short-term events like a flash flood or a wildfire.
What it still gets wrong
DeepMind is open about this being early research. In its own words, Genie 4 can hold a world together over short sessions, but visual details and object permanence can still shift or drift over time, especially in longer runs, after several complex prompt edits, or while you move through a scene. Complex physical interactions, the accuracy of real-time audio, rendering legible text, spatial accuracy and long-term consistency are all listed as active research.
Every video and every piece of audio Genie 4 makes carries SynthID, DeepMind's imperceptible watermark, so it can be identified as AI-generated.
Why it matters
A world model is a step past video generation. A generated clip ends; a generated world keeps answering you. If that works reliably, it changes how films get planned, how games get prototyped and how robots and agents get trained before they touch anything real. It also makes "was this footage real" a harder question, which is why the watermarking matters as much as the demo.
Related on Mishal Brands
- GPT-6.1 Terra: the lower-cost OpenAI model built for computer use
- Runway · Luma · Dreamina, the video tools in our library
- Daily AI news
Sources (checked Oct 28, 2026)
- Google DeepMind, "Genie 4: A new generation of world simulation" (Oct 27, 2026): https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
Disclosure: this page has no affiliate links. No company paid for or reviewed this page. See our affiliate disclosure and how we review.
Read next: All trends · Anthropic's 86% study on AI and expert judgment
Browse by category: All 48 Video tools. Every tool page lists the vendor’s own price and the date we checked it.