🔍 Read the full analysis: A Beginner’s Guide To Google’s Genie 4 And Its AI Worlds on ThorstenMeyerAI.com
Get smart everyday buys 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 introduced Genie 4 on October 27, describing it as a real-time AI model that generates interactive scenes with sound and responds to controls and text prompts. The demos are company-selected, public access has not opened, and DeepMind says scenes can shift over time and support only a few minutes of continuous interaction.
Google DeepMind introduced Genie 4 on October 27, presenting it as a real-time AI model that creates interactive scenes from video-like predictions and responds to keyboard, mouse and text input. The announcement matters because it points toward AI-generated worlds that can change while a person interacts with them, but access remains limited to trusted testers, and DeepMind says scenes can drift and run for only a few minutes.
DeepMind describes Genie 4 as a world model: rather than rendering a scene from a stored 3D map, it predicts what the next frames should look like based on what has appeared so far and on the user’s actions. In the company’s demonstration, a user could move through a scene and type instructions such as changing the weather. The model generates the scene as it goes; it is not a finished video clip or a conventional game world built from a fixed map.
DeepMind says Genie 4 can generate sound alongside visuals, including ambient audio, music and character dialogue, and can respond to instructions while a scene is running. It also describes improved physics and human-like characters. Those capabilities are based on the company’s account and selected demonstrations; the announcement provided no benchmark figures to compare performance with other models. The supplied source notes that audio generation is not unique to Genie 4, citing Runway’s GWM Worlds 2.
DeepMind says Gemini 3.8 Flash can act as a guiding agent, interpreting goals, breaking them into steps and monitoring the generated scene. One robotics demonstration showed a simulated Apptronik Apollo humanoid playing checkers, but no physical robot took part, according to the source material. DeepMind has described film previsualization and game prototyping as tester feedback, while robot training remains a research demonstration.
For now, Genie 4 is being tested by trusted testers. DeepMind says it is coming to Google AI Ultra subscribers through Project Genie and to early Gemini API partners, but has not supplied a launch date or API pricing. The source identifies Google AI Ultra as a $200-a-month subscription tier at the time of publication.
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
Where Interactive AI Could Be Used
If Genie 4 can reliably maintain a scene while people alter it, it could give game developers and filmmakers a faster way to explore layouts, camera moves or visual concepts before building them with conventional tools. DeepMind says testers have reported using it for game prototypes and film previsualization. Those are early use cases, not evidence that the model is ready to replace game engines or production workflows.
World models could also support research into simulated environments for robotics. The Apollo checkers example shows a generated scenario, not a robot learning or performing a task in the physical world. The distinction matters: a research demo is not proof of real-world capability. Any practical use will depend on how well the model preserves object identity, follows instructions and represents physical behavior across longer sessions.
More realistic video combined with matching sound also raises questions about identifying synthetic media. Google says Genie 4’s video and audio outputs carry an invisible SynthID watermark. That is a company statement about the output, not an independent assessment of how reliably the watermark can be detected or retained after editing.
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 (Graphite)
Brother TD-2020A 2‑Inch Direct Thermal Desktop Label Printer with USB
As an affiliate, we earn on qualifying purchases.
How Genie Reached Real-Time Worlds
Genie is part of DeepMind’s progression toward models that generate environments rather than only fixed clips. The first Genie, announced in February 2024, made simple two-dimensional worlds; its research paper described an 11-billion-parameter model trained on unlabeled internet videos. Genie 2 followed in December 2024, generating 3D worlds from a single image and maintaining them for up to a minute, though its real-time version was a lower-quality, reduced model.
In August 2025, DeepMind presented Genie 3 as its first real-time world model, reporting 720p output at 24 frames per second and consistency for a few minutes. Google opened Genie 3 through Project Genie to paying U.S. subscribers in January 2026. TechCrunch reported a 60-second limit per world, citing computing costs. That reported limit is distinct from DeepMind’s wording about Genie 4, which says interaction lasts a few minutes.
The central difference from a conventional game engine is that Genie predicts what the scene should look like next instead of drawing each view from a prebuilt 3D map. That can make an environment flexible, but also creates a risk of visual drift: objects may change or fail to remain consistent when the user returns to them.
“Details and objects “can shift or drift over time.””
— Google DeepMind
Access, Quality and Session Limits
Public availability remains undated. DeepMind has said access is coming to Google AI Ultra subscribers through Project Genie and to early Gemini API partners, but the source material gives no release date, API price or detail on how many users will be admitted. It is also unclear whether the announced access will include all the features shown in the demonstrations.
The public evidence is limited to company-selected clips and descriptions, not independent testing or published benchmark results. DeepMind’s stated session length is “a few minutes”; TechCrunch and Bloomberg, as cited by heise, have reported scenes lasting about two minutes, but that figure is not in DeepMind’s own announcement. The relationship between those descriptions and any limits in the available product is not yet clear.
Long-term consistency, output quality across different prompts, and the model’s usefulness for production or robotics remain unverified. DeepMind itself warns that details can drift. The degree to which SynthID watermarking persists through editing, and how outside parties can check it, is also not established in the supplied material.
The Next Test Is Wider Access
The next concrete milestone is broader access through Project Genie and the Gemini API. Google has not announced dates or pricing for those routes. Until access expands, outside users will have limited ability to evaluate the demonstrations or test how Genie 4 performs on their own tasks.
Further evidence will be needed on how long scenes remain coherent, whether objects and characters stay consistent, and how well audio and interactive instructions work outside curated examples. For developers and researchers, the practical test will be whether the system can support repeatable workflows—not just create convincing short demonstrations. For now, Genie 4 is an announced research and product development, with its capabilities and limits still subject to testing.
Key Questions
What is Google Genie 4?
Genie 4 is a world model announced by Google DeepMind. The company says it generates scenes as users interact, predicting new frames in response to controls and text instructions rather than rendering a fixed video.
Can the public use Genie 4 now?
No public launch date has been given. DeepMind says access is coming to Google AI Ultra subscribers through Project Genie and early Gemini API partners, while current testing is limited to trusted users.
Does Genie 4 generate sound?
DeepMind says it generates audio alongside the scene, including environmental sound, music and dialogue. That capability is the company’s description; the announcement did not provide independent benchmarks.
How long can a Genie 4 world run?
DeepMind says the model supports a few minutes of continuous interaction and warns that details can drift. Reports of scenes lasting about two minutes are attributed to TechCrunch and Bloomberg as cited by heise, not stated in DeepMind’s announcement.
Is Genie 4 already training real robots?
The source describes a robotics research demonstration involving a simulated Apollo humanoid playing checkers. No physical robot was involved in that demo, so it does not establish that Genie 4 is training robots in the real world.
Source: ThorstenMeyerAI.com
