TL;DR
Get everyday essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Google DeepMind presented Genie 4 on October 27, 2026, describing a real-time world model that generates interactive scenes with sound and can respond to keyboard, mouse and text inputs. The demonstrations are company-selected, independent benchmark results were not provided in the source material, and DeepMind says scenes can drift after a few minutes. Access is limited to trusted testers, with broader availability planned but no date announced.
Google DeepMind demonstrated Genie 4 on October 27, presenting it as a real-time AI world model that can generate interactive scenes, sound and responses to typed instructions. The company says trusted testers can use it now, but the public demonstrations do not establish independent performance results, and DeepMind acknowledges that details may drift during interactions lasting a few minutes.
In the examples described by DeepMind, users move through generated scenes with keyboard and mouse controls, then change them by typing instructions such as altering the weather or adding an object. The model generates subsequent frames as the scene changes rather than rendering a completed clip in advance. DeepMind also says the system produces audio alongside the imagery, including environmental sounds, music and dialogue.
These capabilities are DeepMind’s descriptions of its own system, shown in demonstrations selected by the company. The source material reports no independent benchmark figures. It also says DeepMind warns that objects and details “can shift or drift over time” and that the model supports “a few minutes of continuous interaction.” TechCrunch and Bloomberg, as cited by German outlet heise, reportedly described clips of about two minutes; that figure is not in DeepMind’s own wording and should not be treated as a published benchmark.
Availability is limited to trusted testers. DeepMind says access is expected to come to Google AI Ultra subscribers through Project Genie and to early partners using the Gemini API, but it has not given a release date or an API price. The source identifies Google AI Ultra as a $200-per-month subscription tier at the time of publication.
Interactive Worlds Beyond Video
Genie 4 points to a different way of generating digital scenes: rather than asking an AI to make a fixed video, a user can move through an environment and request changes while it is running. If the approach proves reliable, it could give creators a faster way to test visual ideas and give researchers another tool for simulating situations. For now, those possibilities are not the same as demonstrated, broadly available products.
DeepMind says testers have used the system for film previsualization and game prototypes; these are reported as tester feedback, not independent evaluations. The robotics example described in the source is a simulation, not a test with a physical robot. The distinction matters to businesses weighing whether the system is ready for production or research use: the available evidence comes mainly from selected demonstrations and limited testing, while consistency over longer sessions remains unresolved.
The generated audio also broadens the kinds of synthetic media the system can produce. The source says Google claims Genie 4’s video and audio outputs carry an invisible SynthID watermark. That may help with provenance, but the material does not provide details about how the watermark can be detected or how well it survives editing and redistribution.
Popular right now
Shark Navigator Lift-Away Upright Vacuum (NV360)
Vitamix Propel Series 510 Professional-Grade Blender, 3 Programs, Black
KitchenAid Classic Series 4.5 Quart Tilt-Head Stand Mixer K45SS, Onyx Black
As an affiliate, we earn on qualifying purchases.
From Genie 2 to Genie 4
Genie is part of DeepMind’s work on world models: systems that generate what happens next in a scene based on its prior frames and a user’s input. Unlike a conventional video generator, which produces a clip to watch, a world model is intended to respond as someone moves or gives instructions. Unlike a standard video game, the source says Genie does not rely on a stored 3D map of every object and surface; it predicts subsequent frames, a design that may help explain why visual details can drift.
The source traces the series from an early Genie research model described in a 2024 paper, through Genie 2’s generated 3D worlds, to Genie 3, which DeepMind introduced as a real-time world model in 2025. Genie 4’s described additions include more realistic imagery and people, live text-based changes, and audio generated with the scene. The company also describes Gemini as a guiding agent that breaks goals into steps and monitors the resulting frames. Those descriptions outline Google’s account of the system, not a verified independent assessment of its performance.
One application mentioned in the source has an earlier precedent: Waymo uses a simulator built on Genie 3, according to the material. That does not establish that Waymo uses Genie 4. The distinction between a research demonstration, a predecessor used in a simulator and a deployed Genie 4 application remains important.
Performance and Access Remain Open
Several practical questions remain unanswered. The source material gives no independent benchmark results or detailed comparison with other world models. It also does not establish how long a typical scene remains coherent, how performance varies across different prompts, or how frequently generated objects change unexpectedly. DeepMind’s own caveat is that scenes can drift, while other outlets’ reported clip duration is not stated in the company’s quoted wording.
The timing and terms for wider access are also unsettled: no date is given for Google AI Ultra subscribers, no API price is specified, and the number or selection criteria for trusted testers are not provided. The source does not explain whether SynthID can be reliably identified after outputs are edited or shared. Claims about potential uses should be read with similar care: film and game applications are reported tester feedback, while the cited humanoid-robot scenario was simulated and involved no physical robot.
Broader Release Has No Date
The next visible milestone is any expansion of access beyond trusted testers. DeepMind has indicated that Project Genie access for Google AI Ultra subscribers and access for early Gemini API partners are planned, but it has not announced when either will begin. Until Google publishes fuller access terms and technical details, users and organizations cannot assess availability, cost or suitability for sustained work.
Further demonstrations or third-party testing could clarify how consistently Genie 4 follows instructions, preserves scene details and generates sound over longer sessions. For now, the confirmed development is a company presentation and a limited testing program—not general public access or a validated replacement for conventional game engines, video tools or physical robotics testing.
Key Questions
What is Genie 4?
Genie 4 is a world model that Google DeepMind says generates interactive scenes in real time. Users can move through a scene and give instructions that change what it generates next.
Can people use Genie 4 now?
According to the source material, access is limited to trusted testers. Google has not announced a public release date; access for Google AI Ultra subscribers and early Gemini API partners is described as planned.
How long can a Genie 4 scene stay consistent?
DeepMind says the system supports a few minutes of continuous interaction and warns that details can shift or drift. Reports cited by the source describe clips of about two minutes, but that number is not in DeepMind’s own quoted description.
Does Genie 4 control a real robot?
The robotics example in the source is a simulation involving a model and Apptronik’s Apollo humanoid. The source says no physical robot took part in that demonstration.
Has Google announced Genie 4 pricing?
No. The source says Google has not provided an API price or an access date. It identifies Google AI Ultra as a $200-per-month tier at the time of publication, but that is not a stated Genie 4 price.
Source: Thorsten Meyer AI
