🔍 Read the full analysis: What Can Google’s Genie 4 Do—and Where Does It Still Break? 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 a real-time AI model that generates interactive scenes with sound and accepts text changes while running. The demonstrations are company-selected, the model is not yet generally available, and DeepMind says scenes can shift or drift during a few minutes of interaction.
Google DeepMind presented Genie 4 on October 27, describing it as a real-time AI model that generates interactive scenes, sound and responses to typed instructions. The demonstrations suggest a new way to prototype virtual environments, but the system is not yet generally available, and DeepMind says objects and details can drift during sessions lasting only a few minutes.
In DeepMind’s account, Genie 4 predicts what happens next in a scene as a user moves through it with keyboard or mouse controls. A user can also type changes while it runs, such as asking for rain or a camera movement. Unlike a conventional game, the model does not rely on a stored 3D map of the environment; it generates the next frames from what it has seen and the user’s input.
DeepMind says the model produces audio alongside visuals, including ambient sounds, music and dialogue, and simulates events such as collisions and aerodynamics. The company also describes Gemini 3.8 Flash as an agent that can interpret goals, break them into steps and monitor the generated scene. These capability descriptions are vendor claims; DeepMind did not publish benchmark figures in the supplied announcement material.
Access is currently limited to trusted testers. DeepMind says it plans to offer Genie 4 to Google AI Ultra subscribers through Project Genie and to early Gemini API partners, but has not given a release date or API price. Third-party reports cited in the source describe scenes lasting about two minutes; that duration is not specified in DeepMind’s own wording, which says interaction lasts “a few minutes.”
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
A New Tool for Interactive Prototypes
If the demonstrations translate into dependable access, Genie 4 could let creators and developers test ideas in a scene they can alter as it runs, rather than first building a complete environment in a conventional game engine. DeepMind says testers have tried it for film previsualization and game prototyping. Those are reported feedback and early uses, not evidence of established production results.
The robotics example points to a separate research possibility: generating simulated settings for testing tasks. The source describes a simulation of Apptronik’s Apollo humanoid playing checkers, with Gemini helping break the task into steps. No physical robot took part, so the demonstration does not establish that Genie 4 can train a robot for real-world operation. Waymo’s use of predecessor Genie 3 in its driving simulator provides relevant company context, but it does not confirm a Genie 4 deployment.
Sound generated with images also increases the importance of provenance. DeepMind says Genie 4 video and audio outputs carry an invisible SynthID watermark. The announcement material does not establish how reliably that mark can be detected or how it will be used in practice.
Products Worth Considering

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400

LG 27GR83Q-B 27-inch Ultragear QHD IPS Gaming Monitor

Samsung 32 Odyssey G5 QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Short Clips to Live Worlds
Genie 4 follows a series of models with increasing interactivity. The first Genie, introduced in February 2024, generated simple 2D worlds; its research paper described an 11-billion-parameter model trained on unlabeled internet video. Genie 2, presented in December 2024, generated 3D worlds from a single image and maintained consistency for up to a minute, while real-time use required a lower-quality version.
In August 2025, DeepMind described Genie 3 as its first real-time world model, running at 720p and 24 frames per second and maintaining a world for a few minutes, with visual memory reaching back about a minute. Google later opened Genie 3 through Project Genie to paying U.S. subscribers. TechCrunch reported a 60-second limit per world, attributing it partly to computing costs. These earlier limits underline why Genie 4’s stated session length and consistency matter, though the new model’s performance cannot be judged from the announcement alone.
Genie 4’s showcase clips were selected and edited by DeepMind, according to the source material. They illustrate the company’s intended uses, but do not show how the system performs across unscripted prompts, repeated sessions or longer interactions.
“Objects and details “can shift or drift over time.””
— Google DeepMind
Limits Behind the Demonstrations
Several key performance questions remain open. DeepMind has not published benchmark numbers in the supplied announcement, and its selected clips cannot show how often scenes remain stable under varied prompts. The company itself says details and objects may drift, and gives only a broad “few minutes” duration. Reports of sessions lasting about two minutes come from outside DeepMind’s own announcement and should not be treated as a confirmed product limit.
Availability is also unsettled: there is no launch date, confirmed Gemini API price or detailed account of who will qualify as a trusted tester. It is not clear how well the generated sound and visuals remain synchronized across longer sessions, or how reliably the SynthID watermark identifies outputs. The robotics demonstration is a simulation, not a test of a physical machine, and reported creator interest does not establish commercial results.
Tester Access and Public Evidence
The next concrete development will be whether DeepMind opens Genie 4 to the trusted testers it has named and shares further details for Project Genie or Gemini API partners. Google has not announced a timetable or pricing for that access. More hands-on use could clarify how long sessions last, how often scenes drift and whether users can reliably direct changes while a world continues to run.
Until independent users can test the system and results can be compared across prompts and tasks, the announcement remains an account of vendor demonstrations and early feedback. Evidence from accessible tests, published technical details and any measured comparisons will be needed to assess how useful Genie 4 is beyond its showcase clips.
Key Questions
What is Google Genie 4?
DeepMind describes Genie 4 as a real-time world model that generates the next moments of a scene as a user moves through it or types instructions. The source says it does not rely on a conventional stored 3D map.
Can people use Genie 4 now?
No general release is described. DeepMind says access is for trusted testers, with planned availability for Google AI Ultra subscribers through Project Genie and early Gemini API partners. No date or API price was given.
What are Genie 4’s main reported capabilities?
DeepMind says users can steer scenes with keyboard and mouse controls, make changes with text, and generate sound with visuals. The company also claims improved simulation of people and physical events, but has not supplied benchmark numbers in the source material.
What are Genie 4’s known limits?
DeepMind says scenes can shift or drift and supports only a few minutes of continuous interaction. Outside reports cited in the source give an approximate two-minute duration, but that figure is not in DeepMind’s own statement.
Does Genie 4 train real robots?
The source describes a robotics research demonstration conducted in simulation; no physical robot was involved. It does not show that Genie 4 has trained a robot to perform tasks in the real world.
Source: ThorstenMeyerAI.com
