TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Google DeepMind presented Genie 4 on October 27, describing a real-time AI model that generates interactive scenes with sound and accepts text instructions while they run. The demos are company-selected, public access has not begun, and DeepMind says scenes can drift after a few minutes.
Google DeepMind presented Genie 4 on October 27, a new AI “world model” that the company says can generate interactive scenes with synchronized sound and change them in response to text prompts while they run. The announcement matters for game, film and robotics researchers, but the system is not publicly available, and DeepMind acknowledges that scene details can drift during longer sessions.
In DeepMind’s demonstration, a user enters a scene and moves through it with keyboard or mouse controls, then types instructions such as changing the weather or adding an object. The model generates subsequent frames based on the scene and the user’s input. Unlike a conventional game, Genie 4 does not use a stored 3D map and game engine to render its world; it predicts what comes next from what has already been generated. That approach can make a scene feel responsive, but also leaves room for its details to change inconsistently.
DeepMind describes four advances: more realistic visuals and movement, interactive text instructions, audio generated alongside the images, and an associated Gemini model that helps plan and monitor actions. The company says the audio can include ambience, music and dialogue. Those points are vendor descriptions, not independently verified benchmark results: DeepMind published no performance figures in the supplied announcement material, and its own clips were selected demonstrations.
Access is limited to trusted testers for now. DeepMind says Google AI Ultra subscribers will be able to try Genie 4 through Project Genie “soon,” and that early partners will get access through the Gemini API. The company has not provided a release date, API pricing or details on how broad testing will be. DeepMind says outputs will carry an invisible SynthID watermark.
Where Interactive Worlds Could Matter
If the demonstrations hold up in wider testing, Genie 4 could offer a faster way to explore environments before building them with conventional production tools. DeepMind says testers have used the system for film previsualization and game prototypes. These are reported examples of early feedback, not evidence that the model is ready to replace established workflows or ship a finished game.
For robotics, generated environments could help researchers test simulated tasks. The announcement describes an experiment in which Gemini broke a task into steps and Genie 4 simulated Apptronik’s Apollo humanoid playing checkers. No physical robot took part, according to the supplied reporting, so the demo does not establish that the simulation transfers reliably to real-world robot behavior. Waymo already uses the predecessor Genie 3 in its driving simulator, a more concrete example of how the research line may inform simulation work.
The audiovisual output also raises questions about how people identify synthetic media. DeepMind says Genie 4 video and audio carry invisible SynthID watermarks. That is the company’s stated safeguard; the supplied material does not report independent tests of watermark detection or explain how consistently it can identify edited or re-encoded outputs.
Products Worth Considering
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
As an affiliate, we earn on qualifying purchases.
From Genie 3 to Real-Time Scenes
Genie 4 follows a sequence of increasingly interactive models. DeepMind’s first Genie, introduced in February 2024, generated simple 2D worlds from video training data. Genie 2, announced in December 2024, generated 3D environments from a single image and maintained consistency for up to a minute, according to the company. In August 2025, Genie 3 was presented as a real-time model, with DeepMind describing 720p output at 24 frames per second and consistency lasting a few minutes.
The central distinction is between generating a finished video and generating a scene that responds as it unfolds. A video model creates a clip from a prompt; a world model attempts to predict each next frame in response to movement or instructions. Genie 4’s reported ability to add sound and accept live text changes extends that interactive idea. It remains a generative simulation, not a conventional 3D game engine, and that difference helps explain both the flexibility shown in demos and the possibility of visual drift.
““a few minutes of continuous interaction””
— Google DeepMind
Limits Behind the Demo Clips
Genie 4 remains in restricted testing, so its performance outside curated demonstrations is not yet clear. DeepMind has not published benchmark numbers in the supplied material, and there is no independent evaluation of the model’s visual consistency, sound quality or response to user instructions. Reports cited in the source material describe some scenes lasting about two minutes, while DeepMind’s own wording says “a few minutes”; the company has not provided a more precise public duration.
It is also unclear how often objects change unexpectedly, whether the system can maintain complex scenes across longer sessions, or what hardware and latency are required. The robotics example is a simulation rather than a physical-robot test. The planned tester rollout has no date, and pricing and access conditions for Gemini API partners have not been specified. Claims about realism, usefulness and reliability therefore remain provisional until more users and independent observers can examine the system.
Waiting for Tester Access
The next concrete step is access for trusted testers, followed by planned availability for Google AI Ultra subscribers through Project Genie and early Gemini API partners. DeepMind has not announced dates for either stage. Details to watch include whether users can make their own scenes, the limits on session length, API terms and cost, and whether the model’s performance differs from the selected launch clips.
Broader testing should also show whether the claimed gains in interactive audio and scene behavior are useful beyond short demonstrations. Until Google releases access and more technical detail, Genie 4 is best understood as a research and product announcement—not a publicly testable tool with independently established capabilities.
Key Questions
What is Google Genie 4?
Genie 4 is a world model that Google DeepMind says generates interactive scenes frame by frame in response to movement and text prompts. It is not a conventional game built around a stored 3D map.
Can the public use Genie 4 now?
No. DeepMind says access will begin with trusted testers, with planned access for Google AI Ultra subscribers through Project Genie and early Gemini API partners. No dates have been announced.
What can users do in a Genie 4 world?
In DeepMind’s demonstrations, users move through scenes with keyboard or mouse controls and type instructions to change elements such as weather or objects. The company also says the model generates audio alongside the visuals.
What are Genie 4’s known limitations?
DeepMind says details and objects can shift or drift over time and describes sessions as lasting “a few minutes.” Independent performance tests and more precise duration figures have not been supplied.
Has Genie 4 been tested with a real robot?
The supplied account describes a robotics demonstration in simulation, including a simulated Apptronik Apollo playing checkers. No physical robot was involved in that example.
Source: Thorsten Meyer AI
