🔍 Read the full analysis: A Closer Look At What Google’s Genie 4 Can Build on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google DeepMind presented Genie 4 on October 27, describing it as a real-time world model that generates interactive scenes, sound and changes prompted by text. The demonstrations are company-selected, access is limited to trusted testers, and DeepMind says scenes can shift or drift after a few minutes.
Google DeepMind demonstrated Genie 4 on October 27, showing an AI-generated scene that users can explore and alter with keyboard controls and text prompts. The company says the model creates visuals and audio together and can respond while a scene is running, but its demonstrations are selected by DeepMind, access remains limited, and the company acknowledges that scenes can shift or drift over time.
DeepMind describes Genie 4 as a real-time world model: rather than producing a finished video from a prompt, it generates successive frames in response to what a user does. In the company’s demonstration, a person moves through a scene with keyboard controls and enters instructions to change conditions such as the weather. DeepMind says users can also prompt changes to the camera, people or objects.
The company describes four main capabilities: interactive control, more realistic people and physical behavior, audio generated alongside the scene, and support from a Gemini model that can break goals into steps and monitor results. DeepMind says the system can generate environmental sounds, music and character dialogue. These descriptions are vendor claims; the source material reports no independent benchmark results.
DeepMind says Genie 4 is available to trusted testers, with future access planned for Google AI Ultra subscribers through Project Genie and for early Gemini API partners. It has not announced a release date or API pricing. DeepMind says sessions can run for a few minutes; TechCrunch and Bloomberg, as cited by heise, reported scenes lasting about two minutes, a figure not included in DeepMind’s own 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
From Generated Clips to Interactive Scenes
Genie 4 points toward a different way to use generative AI: people may be able to explore a scene as it is created, rather than simply request a clip and watch the result. That could be useful for game prototyping and film previsualization, where teams need to test environments, movement or visual ideas quickly. DeepMind has cited tester feedback on those uses, but the reports do not establish how well the tool performs in production.
The distinction also matters for robotics research. A generated environment can let researchers test simulated situations, but a convincing simulation is not the same as a real robot acting in the physical world. The demonstration described in the source material used a simulated Apptronik Apollo robot playing checkers; no physical robot was involved. Any claims about practical robotics value will depend on how well simulated scenes reflect real conditions.
There is also a verification issue as generated scenes become more lifelike and include matching sound. DeepMind says Genie 4’s video and audio outputs carry an invisible SynthID watermark. That is the company’s stated safeguard; the material does not provide independent evidence about how reliably the watermark can be detected or retained when media is edited.
Products Worth Considering
Logitech MX Mechanical Wireless Illuminated Keyboard Tactile (Graphite)
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Genie’s Move Toward Real Time
Genie 4 follows a series of DeepMind models that moved from basic generated environments toward interactive scenes. The first Genie, introduced in February 2024, produced simple 2D worlds. Genie 2, announced in December 2024, generated 3D worlds from a single image and could maintain them for up to a minute, though real-time operation used a reduced-quality version.
In August 2025, Genie 3 added real-time generation, with DeepMind reporting 720p video at 24 frames per second and consistency for a few minutes. Google later made Genie 3 available to paying US subscribers through Project Genie, according to the source material. That history offers a useful comparison, but it does not establish that Genie 4’s capabilities or planned availability will match its demonstrations.
Unlike a conventional game engine, which uses a stored 3D map to render a scene, DeepMind says Genie 4 predicts what comes next from the frames it has generated and a user’s input. This can support flexible interaction, but it may also contribute to continuity problems: objects or details may change as a user moves through a scene.
“Objects “can shift or drift over time.””
— Google DeepMind
Access, Performance and Drift
Several practical details remain unsettled. DeepMind has not said when wider access will begin, what the Gemini API will cost, or how much control users will have outside the selected demonstrations. The source material also does not provide independent testing, benchmark scores or a detailed account of how performance varies across prompts and scene types.
Session length is another open point. DeepMind says interaction lasts a few minutes, while TechCrunch and Bloomberg, as cited by heise, report scenes of about two minutes. The reports do not establish whether that duration applies to all users or configurations. DeepMind itself says details and objects can drift over time, and it has not supplied a quantified rate or a clear description of how the issue affects different uses.
Claims about applications should also be treated with care. Film and game uses are described as tester feedback reported by DeepMind, while the robotics example is a research simulation rather than a test with a real robot. The available material does not show whether Genie 4 can meet professional reliability requirements or reduce costs in those settings.
Waiting for Wider Access
The next public milestone is a fuller rollout, which DeepMind says is planned for Google AI Ultra subscribers through Project Genie and early Gemini API partners. The company has not announced dates, API pricing or the limits that subscribers will face, so those details will be needed to judge who can use the system and for what kinds of work.
Independent access and testing will also help clarify how Genie 4 performs beyond the company’s chosen clips. Useful evidence would include measured session duration, scene consistency, audio quality, responsiveness to prompts and the reliability of SynthID on generated outputs. Until then, the clearest account is that Genie 4 has been demonstrated, not broadly released, and its advertised strengths and limitations remain based chiefly on DeepMind’s description and demonstrations.
Key Questions
What is Google DeepMind’s Genie 4?
It is a world model that DeepMind says generates successive scene frames in response to user controls and prompts, rather than rendering a completed video in advance.
Can users try Genie 4 now?
DeepMind says it is available to trusted testers. Wider access through Project Genie for Google AI Ultra subscribers and through early Gemini API partners is described as coming soon, but no date has been announced.
What can Genie 4 generate?
According to DeepMind, it can create interactive visual scenes, generate audio such as environmental sounds and dialogue, and respond to text instructions that change elements of a running scene. These capabilities have been presented in company-selected demonstrations.
How long can a Genie 4 scene run?
DeepMind says the model supports a few minutes of continuous interaction. Some media reports cited in the source material describe scenes lasting about two minutes, but DeepMind has not stated that figure in its own description.
Does the robotics demo show a real robot using Genie 4?
No. The described checkers demonstration simulated an Apptronik Apollo humanoid; the source material says no physical robot was involved.
Source: ThorstenMeyerAI.com
