🔍 Read the full analysis: Can Google’s Genie 4 Make AI Worlds Feel Interactive? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google DeepMind demonstrated Genie 4 on October 27, describing a system that generates interactive scenes as users move and give text instructions. The company says it adds synchronized audio and more realistic environments, but it has published no benchmark figures, acknowledged that scenes can drift after a few minutes, and has not opened access beyond trusted testers.
Google DeepMind demonstrated Genie 4 on October 27, showing an AI system that generates a scene as users move through it and enter new instructions. The company says the world model can produce visuals and sound together and respond to changes such as weather, but the system is not yet generally available and DeepMind says scenes can drift during longer sessions.
In DeepMind’s demonstrations, users steer generated environments with keyboard and mouse controls, then type instructions to change what is happening. The company says the system can add conditions such as rain, alter the camera, or introduce people and objects while a scene continues to run. Unlike a conventional game built around a stored 3D map, a world model generates what comes next from the preceding frames and user input.
DeepMind describes Genie 4 as its first real-time world model to generate audio natively, including environmental sounds, music and character dialogue. It also says Gemini 3.8 Flash can interpret goals, break them into steps and guide the generated scene. These are company descriptions and selected demonstrations; DeepMind did not provide benchmark numbers in the supplied announcement material.
Access is currently restricted to trusted testers. DeepMind said it plans to make Genie 4 available to Google AI Ultra subscribers through Project Genie and to early Gemini API partners, but gave no release date or API pricing. DeepMind’s own description says objects and details can “shift or drift over time” and that the system supports “a few minutes of continuous interaction.” TechCrunch and Bloomberg, as cited by heise, report clips of up to about two minutes; that figure was not stated in DeepMind’s description.
Infographic 01 · How Genie grew up
From flat sketches to worlds with sound, in under three years.
Each step as Google DeepMind described it. Session limits and consistency are different measures, so read the labels.
- Feb 2024
Genie 1Simple 2D worlds11 billion parameters, learned from unlabelled internet videos - Dec 2024
Genie 23D worlds from a single imageConsistent “for up to a minute”; real time only as a slimmed-down, lower-quality version - Aug 2025
Genie 3First real-time Genie720p at 24 frames per second, consistent “for a few minutes”, memory about one minute back - Jan 2026
Project GenieUS AI Ultra subscribers can play60 seconds per world, partly because of compute costs - May 2026Project Genie goes globalIn the $200 AI Ultra tier, for users 18 and older
- 27 Oct 2026
Genie 4Sound, live steering, more realismResearch preview for trusted testers; Ultra subscribers and API partners “coming soon”
Genie 3 vs. Genie 4
As described by Google DeepMind. Genie 4 column = vendor claims unless marked.
| Feature | Genie 3 (Aug 2025) | Genie 4 (Oct 2026) |
|---|---|---|
| Sound | Genie 3: None | Genie 4: Generated with the picture: ambient noise, music, dialogue |
| Steering | Genie 3: Keys, plus “promptable world events” | Genie 4: Free-form text changes while it runs; Gemini adapts and checks them |
| Look | Genie 3: 720p, 24 fps | Genie 4: “Unprecedented realism”; no resolution or frame rate published |
| How long | Genie 3: “A few minutes”; Project Genie capped at 60 s | Genie 4: “A few minutes” (DeepMind); up to ~2 min consistent scene reported |
| Access | Genie 3: Small preview, then Project Genie | Genie 4: Trusted testers; Ultra and API partners “coming soon” |
Sources: Genie paper, arXiv, 23 Feb 2024; Genie 2 blog, 4 Dec 2024; Genie 3 blog, 5 Aug 2025; TechCrunch, 29 Jan 2026; Google, 19 May 2026; Google DeepMind, Genie 4 blog, 27 Oct 2026; Genie 4 model page (limitations, disclaimer); TechCrunch, 27 Oct 2026. Table wording: Thorsten Meyer.
Powered by Thorsten Meyer AI
Interactive Scenes Beyond Games
If the demonstrations hold up in wider testing, Genie 4 could let people explore and revise generated scenes rather than simply watch a finished clip. That distinction may matter to game prototyping and film previsualization, where creators could test settings or actions through interaction. DeepMind says those uses have come up in feedback from testers; they are not evidence of established commercial results.
The approach also has possible research applications. DeepMind showed a robotics simulation in which Gemini 4 Argon divided a task into steps and Genie 4 simulated Apptronik’s Apollo humanoid playing checkers. The supplied account says no physical robot took part in that demonstration, so it should be understood as a simulation, not a report of a robot completing the task in the real world. Waymo, meanwhile, already uses the predecessor Genie 3 in its driving simulator, according to the source material.
For audiences, increasingly realistic synthetic video and sound raise questions about provenance. DeepMind says Genie 4’s video and audio outputs carry an invisible SynthID watermark. The announcement does not establish how reliably that marking can be detected after content is edited or shared, or how widely the tool will be used once access expands.
Popular right now
Dyson Ball Animal 3 Corded Upright Vacuum
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Philips 3300 Series Fully Automatic Espresso Machine with LatteGo Milk System
As an affiliate, we earn on qualifying purchases.
From Genie 3 to Genie 4
Genie 4 follows a series of systems that DeepMind has presented as steps toward interactive generated environments. The first Genie, introduced in February 2024, created simple 2D worlds. Genie 2, announced in December 2024, generated 3D environments from a single image and maintained them for up to a minute, though its real-time version ran at lower quality, according to the source material.
In August 2025, DeepMind introduced Genie 3 as its first real-time world model, reporting 720p output at 24 frames per second and consistency for a few minutes. Google later offered Genie 3 through Project Genie to paying U.S. subscribers in January 2026. TechCrunch reported a 60-second limit per world, attributed partly to computing costs. Those earlier figures provide a point of comparison, but DeepMind has not supplied equivalent benchmark data for Genie 4 in the material provided.
The core difference from a conventional game engine is that Genie 4 does not use a stored, complete 3D map to render each view. It predicts subsequent frames based on what it has generated and the user’s actions. That design can produce varied scenes, but it also helps explain the company’s caveat that details may drift over time.
““can shift or drift over time””
— Google DeepMind
Limits Behind the Demonstrations
Genie 4 has not been made available for independent public testing, and the demonstrations were selected by DeepMind. The supplied material gives no benchmark results for visual quality, response speed, audio performance or consistency, making it difficult to compare the system with other tools. It is also unclear how much the system’s performance changes across different prompts and environments.
Duration is another open question. DeepMind describes a few minutes of continuous interaction, while outside reports cited in the source material put demonstrated scenes at about two minutes. The reports’ figure is not in DeepMind’s own wording, and there is no detailed public measurement here of how scene stability changes during a session. Pricing, availability dates and API limits are also unannounced.
The possible uses should be treated with similar care. Tester feedback about film and game workflows is not evidence that businesses have adopted the tool at scale. The robotics example was a simulation, and the source does not establish whether Genie 4 has been validated as a practical substitute for physical testing.
Broader Access Has No Date
The next reported step is access for Google AI Ultra subscribers through Project Genie and for early Gemini API partners. DeepMind has not announced when that access will begin, what the API will cost, or what usage limits will apply. Until then, the system’s capabilities remain represented chiefly by company-selected demonstrations and limited tester feedback.
Further details that would help evaluate the announcement include public access for independent users, repeatable performance measurements, and information about how long scenes remain coherent under varied interactions. Google’s rollout information and any outside testing will show whether Genie 4’s interactive visuals and synchronized audio can be used reliably beyond short demonstrations.
Key Questions
What is Google Genie 4?
Genie 4 is a world model that generates visual scenes frame by frame in response to user movement and instructions. It is not described as a conventional game engine with a complete stored 3D map.
Can the public use Genie 4 now?
No. The source material says access is limited to trusted testers. Google has said access is coming to AI Ultra subscribers through Project Genie and early Gemini API partners, without giving a date.
What is new in Genie 4?
DeepMind says Genie 4 combines real-time scene generation with native audio generation, text-based changes while a scene runs, and guidance from Gemini. These capabilities have not been accompanied by public benchmark figures in the supplied material.
How long can a Genie 4 world stay consistent?
DeepMind describes a few minutes of continuous interaction and warns that objects and details can drift. Outside reports cited in the source material say scenes last about two minutes, but that number is not in DeepMind’s own description.
Is Genie 4 being used to train real robots?
The cited robotics example was a simulation of Apptronik’s Apollo humanoid playing checkers; the source says no physical robot was involved. The material does not establish that Genie 4 is currently used to train real robots.
Source: ThorstenMeyerAI.com
