🔍 Read the full analysis: Google’s Genie 4: AI Worlds, Current Limits, And Who Gets Access 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 demonstrated Genie 4 on October 27, describing it as a real-time model that generates interactive scenes with video and audio. The company says it can change scenes in response to typed instructions, but acknowledges that details may drift and that interaction lasts only a few minutes. Access is limited to trusted testers, with broader access promised but not dated.
Google DeepMind demonstrated Genie 4 on October 27, presenting an AI model that generates a scene as a user moves through it and types instructions. The company says the model can produce video and audio together, but it is not yet generally available: access is limited to trusted testers, and DeepMind says scenes may drift during interactions lasting only a few minutes.
Unlike a conventional video generator that renders a finished clip, Genie 4 predicts what happens next as a user supplies keyboard or mouse input. DeepMind says people can also type instructions to change the scene while it is running, such as altering the weather or adding an object. The model does not use a stored 3D map or a conventional game engine, according to the source material; it generates each next moment from what has already appeared and the user’s input.
DeepMind describes the system as supporting more realistic people and physics, including collisions and aerodynamics, and generating ambient sound, music and character dialogue alongside visuals. It says Gemini 3.8 Flash can interpret goals, break them into steps and monitor the generated scene. These capabilities were shown in company-selected demonstrations; DeepMind did not publish benchmark figures in the supplied material, so the claims have not been independently quantified here.
For access, DeepMind says Genie 4 is available to trusted testers. It said the model is coming to Google AI Ultra subscribers through Project Genie and to early Gemini API partners, but gave no release date or API pricing. DeepMind also says its video and audio outputs carry an invisible SynthID watermark.
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
What Interactive Worlds Could Enable
Genie 4 points toward AI systems that can generate environments people interact with, rather than only producing a fixed image or video. If the model’s demonstrated controls and scene changes hold up beyond curated examples, they could make it faster to create game prototypes and film previsualizations. DeepMind has reported tester feedback describing those uses; that is not evidence that the model is ready for routine production.
The announcement also matters to robotics research. The source describes a demonstration in which Gemini divided a task into steps and Genie 4 simulated Apptronik’s Apollo humanoid playing checkers. No physical robot took part, according to the cited account, so the clip shows a simulated scenario, not a successful real-world robot task. Google’s broader work on synthetic environments may interest developers, but the quality and usefulness of those simulations remain open questions.
For viewers, scenes that combine generated images and sound may become more convincing. DeepMind’s stated use of SynthID is relevant to identifying outputs, though the supplied material does not establish how reliably the watermark can be detected after content is edited or shared.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 24GB Memory, 1TB SSD, Midnight
Samsung 49-Inch Odyssey OLED G9 (G91SD) Dual QHD Curved Gaming Monitor
LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Genie 3 to Genie 4
DeepMind’s earlier Genie models provide a short timeline for the announcement. The first Genie, introduced in February 2024, generated simple 2D worlds. Genie 2 followed in December 2024, creating 3D worlds from a single image and maintaining them for up to a minute, though real-time use required a lower-quality version.
In August 2025, Genie 3 was presented as a real-time world model, with DeepMind reporting 720p output at 24 frames per second and consistency for a few minutes. Google later made Genie 3 available to paying U.S. subscribers through Project Genie, according to the supplied account. Waymo uses the predecessor in its driving-simulation work, the source material says. That existing use does not show that Genie 4 is currently deployed in Waymo’s simulator.
Genie 4’s stated additions are native audio generation, live text-directed changes and more natural-looking people and physics. The description of audio as a first is specific to DeepMind’s own real-time world models, not necessarily the wider industry: the source cites TechCrunch as noting another system, Runway’s GWM Worlds 2, also generates audio.
Limits and Access Still Unsettled
It is not clear when Genie 4 will reach Google AI Ultra subscribers or Gemini API partners, what API access will cost, or whether availability will be limited by country, usage or other conditions. The company has not provided a public release schedule in the supplied material. Current access remains restricted to trusted testers.
DeepMind acknowledges that objects and details can drift and that continuous interaction lasts only a few minutes. TechCrunch and Bloomberg, as cited by heise, reported scenes of up to about two minutes; that duration is not stated in DeepMind’s own announcement, according to the source. No public benchmark figures are provided to compare Genie 4’s visual quality, physical consistency, latency or audio with other models.
The demonstrations were selected by DeepMind, and independent access to test the system is not described. The source also characterizes film and game applications as tester feedback and the Apollo example as a research simulation. It remains unclear how well the system performs on varied tasks, how reliably it maintains a scene over time, and whether its generated worlds can support practical work beyond short demonstrations.
Wider Testing Is the Next Test
The next stated step is access for Google AI Ultra subscribers through Project Genie and for early Gemini API partners. DeepMind has not announced when that rollout will begin or supplied pricing for API use. Until those details and wider testing become available, the public evidence remains the company’s demonstrations and its descriptions of trusted testers’ feedback.
Readers and potential users will be looking for independent tests of how long scenes remain coherent, how well typed changes work, and whether sound stays synchronized with the generated action. Further information on availability, performance and limits would help establish whether Genie 4 is mainly a research demonstration or a practical tool for creators and developers.
Key Questions
What is Google Genie 4?
Genie 4 is a Google DeepMind interactive world model. It generates successive moments in a scene as a user moves through it or supplies instructions, rather than rendering only a finished video clip.
Can the public use Genie 4 now?
No general public release is described. DeepMind says access is currently limited to trusted testers, with access for Google AI Ultra subscribers through Project Genie and early Gemini API partners planned for later. No date was given.
What are Genie 4’s current limitations?
DeepMind says details and objects can shift or drift over time and describes continuous interaction lasting a few minutes. The supplied material includes media reports of sessions lasting about two minutes, but says that duration is not in DeepMind’s own announcement.
Does Genie 4 generate sound as well as visuals?
DeepMind says Genie 4 generates audio alongside visuals, including ambient sound, music and dialogue. This is a company-described capability; the supplied material does not include independent performance tests.
Was a real robot used in the Apollo demonstration?
No. The source describes the Apollo humanoid playing checkers in a simulation and says no physical robot was involved. The demonstration should not be read as proof of a real-world robot completing the task.
Source: ThorstenMeyerAI.com
