🔍 Read the full analysis: How Google’s Genie 4 Builds AI Worlds You Can Walk Into on ThorstenMeyerAI.com
Get movie nights delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Google DeepMind presented Genie 4 on October 27, describing it as a real-time AI model that generates interactive scenes with sound and accepts instructions while they run. The company has not released it publicly or provided benchmark data; its own description says scenes can drift and support only a few minutes of continuous interaction.
Google DeepMind presented Genie 4 on October 27, describing an AI system that generates interactive visual worlds, sound and changes prompted by users while a scene is running. The model is available only to trusted testers for now, and DeepMind says its worlds can drift and last only a few minutes—limits that leave its practical uses and public availability unresolved.
In a demonstration described in the source report, a user enters a prompt about a misty lake, a saxophone player and a floating platform, then moves through the scene using keyboard controls. A typed instruction such as “let it rain” changes the weather, while the generated scene includes sounds such as music, wind and water. These examples come from DeepMind’s selected demonstrations, not independent testing.
DeepMind says Genie 4 can respond to keyboard and mouse input, accept text instructions that change a running scene, and generate audio alongside visuals. The company also describes improvements to people and simulated physics, including collisions and aerodynamics. It has published no benchmark figures in the material cited here, so the claims about realism and capability cannot be compared against a stated standardized test.
According to DeepMind, Gemini 3.8 Flash works alongside the model to interpret goals, divide them into steps and monitor generated frames. The announcement also describes a robotics-related demonstration in which a task was simulated with an Apptronik Apollo humanoid. The source report says no physical robot took part. DeepMind identifies trusted testers as the current users; it says access is planned later for Google AI Ultra subscribers through Project Genie and for early Gemini API partners, without giving dates or API pricing.
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
From Generated Clips to Interactive Scenes
Genie 4’s announced features point to a shift from generating a finished video to producing a scene that responds as a user moves through it. If the system works as described, users could alter a setting during a session rather than restarting with a new prompt. That may be useful for game prototyping and film previsualization, both of which DeepMind says testers have tried, though those uses are reported as tester feedback rather than independently verified results.
The distinction also matters for companies evaluating AI-generated environments. A system that maintains a scene as it changes could support interactive demonstrations or research simulations, but short sessions and visual drift may limit tasks that need repeatable, stable conditions. The robotics example is a simulation demonstration, not evidence that Genie 4 has been validated for training real robots.
Generated audio and imagery also raise questions about how audiences identify synthetic media. The source report says Google will attach an invisible SynthID watermark to Genie 4 video and audio outputs. That is Google’s stated approach; the material provided does not establish how reliably the watermark can be detected or retained after editing.
Popular right now
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Ninja SP151 Flip Toaster Oven and Air Fryer, 8-in-1, Stainless Steel
As an affiliate, we earn on qualifying purchases.
How Genie Has Developed
Genie follows earlier DeepMind systems. The original Genie, introduced in February 2024, generated simple 2D worlds; its research paper described an 11-billion-parameter model trained on unlabeled internet videos. Genie 2, shown in December 2024, generated 3D worlds from a single image and maintained them for up to a minute, although real-time operation required a reduced-quality version.
In August 2025, DeepMind presented Genie 3 as its first real-time world model, reporting output at 720p and 24 frames per second and consistency for a few minutes. Google later offered Genie 3 through Project Genie to paying U.S. subscribers, according to the source report. TechCrunch reported a 60-second limit per world, which it linked partly to computing costs; that duration is a media-reported figure, not a detail given in DeepMind’s Genie 4 description.
A world model differs from a conventional game engine, which uses a stored map and rules to render a scene, and from a video generator that produces a clip to watch. In DeepMind’s account, Genie generates the next frames from the scene so far and from user input. That approach can make interaction possible without a conventional stored 3D map, but it also means scene consistency is a central challenge.
Access, Consistency and Testing Gaps
Genie 4 is not publicly available in the information provided. Google has not announced a release date, the price of Gemini API access or detailed eligibility requirements for testers. The claims about improved realism, physics and interactive audio are based on DeepMind’s descriptions and chosen clips; independent evaluations and benchmark results are not included in the source material.
There is also no precise session limit from DeepMind beyond “a few minutes.” TechCrunch and Bloomberg, as cited by German outlet heise, reportedly describe scenes lasting about two minutes, but that figure is media-reported, not confirmed in DeepMind’s announcement. It remains unclear how often objects drift, how performance varies across prompts, and whether generated worlds can support dependable work beyond short demonstrations.
The source report says Waymo uses Genie 3, a predecessor, in its driving simulator. That does not confirm that Waymo uses Genie 4 or that the new model is ready for autonomous-driving development. The robotics clip likewise shows a simulated task, not a real-world robot test.
A Wider Test Is Still Pending
The next stated 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 given a timetable, API price or release criteria, so those plans do not establish when most users or developers will be able to try the system.
Public access and third-party testing will help clarify whether Genie 4 can keep environments coherent beyond the company’s selected clips, how its audio and controls perform across different prompts, and whether its short-session limit changes in practice. Until those details emerge, the announcement establishes what Google says the model can do, not how consistently it will work in everyday use.
Key Questions
What is Google Genie 4?
Genie 4 is a Google DeepMind world model that the company says generates interactive scenes in response to user controls and text prompts, with audio generated alongside the visuals.
Can the public use Genie 4 now?
No public release is described. DeepMind says access is limited to trusted testers, with later availability planned for Google AI Ultra subscribers through Project Genie and early Gemini API partners. No dates have been given.
How long can a Genie 4 world run?
DeepMind says the model supports a few minutes of continuous interaction and warns that details and objects can drift. Reports of sessions lasting about two minutes are attributed to media coverage, not DeepMind’s own stated figure.
Does Genie 4 control real robots?
The cited robotics example is a simulation involving an Apptronik Apollo humanoid. The source report says no physical robot was involved, and it does not establish that Genie 4 has been validated for real-world robot training.
Has Google independently demonstrated Genie 4’s performance?
The material describes demonstrations selected by DeepMind and the company’s capability claims, but gives no benchmark results or independent evaluation. Wider testing is needed to assess consistency and performance.
Source: ThorstenMeyerAI.com
