TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Google DeepMind announced Genie 4 on October 27, 2026, describing it as a real-time world model that generates interactive scenes with sound. Access is limited to trusted testers, with planned availability for Google AI Ultra subscribers through Project Genie and early Gemini API partners. DeepMind says scenes can drift and interaction lasts only a few minutes; it has not published benchmark results or a public release date.
Google DeepMind announced Genie 4 on October 27, describing it as a real-time AI model that generates interactive worlds from user inputs, including keyboard controls and text instructions. The company says access is limited to trusted testers, with plans to bring it to Google AI Ultra subscribers through Project Genie and to early Gemini API partners; no public release date has been announced.
DeepMind says Genie 4 generates scenes frame by frame as users move through them, rather than rendering a fixed video or drawing from a stored 3D game map. In its demonstrations, users control movement with a keyboard or mouse and can type instructions to change conditions such as the weather or to add objects. The company says the model also generates sound alongside images, including environmental audio, music and dialogue.
The capabilities are presented in company-selected demonstrations, and DeepMind has not published benchmark figures in the supplied announcement material. The company says Gemini 3.8 Flash can interpret goals, break them into steps and monitor the generated scene. DeepMind also describes more realistic people and physics, but those remain vendor claims rather than independently measured results.
For access, DeepMind says Genie 4 is available to trusted testers. It lists Google AI Ultra subscribers using Project Genie and early Gemini API partners as planned upcoming audiences. The source identifies Google AI Ultra as a $200-per-month tier, but does not give an API price or a date for either expansion. Reporting by TechCrunch and Bloomberg, as cited by heise, describes scenes lasting about two minutes; that duration is not specified in DeepMind’s own text.
Access · Capabilities · Limits
Who Can Use Google Genie 4, And What Can They Build With It?
Google DeepMind says Genie 4 creates interactive worlds in real time, with images and sound. Access starts with trusted testers; wider availability is planned, while release timing and performance remain unconfirmed.
“A few minutes of continuous interaction”
Google DeepMind · stated limitationAt a glance
What is known so far
Announced capabilities and access plans come from DeepMind. Independent benchmarks and a public launch date have not been supplied.
How it works
A scene that responds as you move
DeepMind describes frame-by-frame scene generation without a stored 3D map. That can enable flexible interaction, while details may shift over time.
Steer and instruct
Demonstrations use keyboard or mouse controls. Text instructions can change conditions, such as weather, or add objects.
Build the next moment
The model generates the continuing scene as users move through it, rather than producing only a fixed video or rendering a defined game map.
Images with sound
DeepMind says outputs can include environmental audio, music and dialogue. Video and audio outputs carry an invisible SynthID watermark.
| Format | Scene behavior | Interaction | What the source says |
|---|---|---|---|
| Genie 4 | Generated as the scene continues | Live controls + text | Scene details may drift; sessions last a few minutes |
| Conventional video generator | Produces a clip to watch | Not described as live | Different output format from an ongoing interactive world |
| Game engine | Renders from a defined environment | Player input | Genie 4 is described as having no stored 3D map |
Who gets access?
First testers, then planned audiences
Both wider access routes are described as upcoming. DeepMind has not provided dates, API pricing, or a list of partners.
Trusted testers
Initial access is restricted to selected testers, rather than the general public.
Google AI Ultra
DeepMind plans access through Project Genie. The source identifies Ultra as a $200-per-month tier.
Early Gemini API partners
Selected partners are expected to get access. Broader API availability and pricing remain unspecified.
General public
No public release date or universal access plan has been announced in the supplied material.
Important distinction: “Coming soon” does not establish when access begins, who qualifies, or whether the demonstrations represent typical results.
What could people build?
Promising directions, early evidence
Examples come from company demonstrations and tester feedback. They point to possible uses, not proven production readiness.
Previsualization
Testers cited by DeepMind described using generated scenes to explore shots and visual ideas before production.
Prototype worlds
Interactive scenes may help developers try environments and concepts without building every setting as a conventional game.
Research demonstrations
A humanoid playing checkers was simulated; no physical robot took part. Waymo’s reported simulator use involves Genie 3, not Genie 4.
From research to interaction
The Genie timeline
Each generation moved toward more persistent, real-time environments. Genie 4 adds native audio and live text-based changes, according to DeepMind.
Genie
Research paper described an 11-billion-parameter model trained on unlabeled internet videos.
Genie 2
Generated 3D worlds from a single image and maintained them for up to a minute; real-time use required lower quality.
Genie 3
DeepMind called it its first real-time world model: 720p at 24 frames per second, with consistency for a few minutes.
Genie 4
Announced with native audio generation and live text-based changes to interactive scenes.
“Can shift or drift over time”
DeepMind · scene consistency“Our first real-time world model to generate audio natively”
DeepMind · audio capabilityAudio generation is not unique to Genie 4
TechCrunch also reported audio in Runway GWM Worlds 2Limits & open questions
What has not been established
Company-selected demonstrations describe possible capabilities. Independent measurement and broader product terms are still missing.
Scenes can drift
DeepMind says objects and details may shift. It describes continuous interaction as lasting a few minutes.
No published benchmarks
The supplied announcement material includes no benchmark figures or independent testing. More realistic people and physics remain company claims.
Key terms remain unknown
No public launch date, API price, partner list, or practical SynthID detection details are specified.
Duration note: TechCrunch and Bloomberg, as cited by heise, report scenes lasting about two minutes. That specific figure does not appear in DeepMind’s own announcement text.
Key questions
Quick answers
Who can use Genie 4 now?
Google DeepMind says access is limited to trusted testers. It has not announced general public availability.
Who may get access next?
Google AI Ultra subscribers through Project Genie and early Gemini API partners are planned audiences. No date or API price has been announced.
What can people build?
Demonstrations show steerable, editable worlds. Tester feedback points to film previsualization and game prototypes as early possibilities.
Can Genie 4 train robots?
The reported humanoid checkers scene was a simulation, with no physical robot. Waymo’s cited use is with predecessor Genie 3.
What are the known limits?
DeepMind says scenes can drift and interaction lasts a few minutes. The supplied material has no published benchmarks.
What should users watch for?
A dated access announcement, product terms, API pricing, and tests beyond company-selected demonstrations.
Who Could Use Genie 4 First
The access plan puts Genie 4 first in the hands of selected testers and partners, not the general public. If the planned expansion proceeds, subscribers and developers with Gemini API access could explore interactive scene generation without building every environment as a conventional game or simulation. The current material does not establish how widely the API will be offered or what it will cost.
Potential uses named in the source include film previsualization and game prototyping, based on tester feedback reported by DeepMind. These are early examples, not proof that the model is production-ready. DeepMind’s robotics demonstration simulated a humanoid playing checkers; according to the source, no physical robot took part. A separate, established example is Waymo’s use of the predecessor Genie 3 in its driving simulator, which does not establish that Waymo uses Genie 4.
Generating audio and images together could make synthetic scenes more useful for planning and interactive experiences. It also makes provenance relevant: Google says Genie 4’s video and audio outputs carry an invisible SynthID watermark. The source does not describe how reliably that mark can be detected or how it will appear to people who encounter generated media.
Products Worth Considering
Logitech MX Keys S Wireless Keyboard, Graphite
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
From Genie 3 to Interactive Worlds
The Genie models have progressed from simple generated environments toward real-time interaction. The source dates the first Genie to February 2024 and says 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 real-time use required a lower-quality version.
In August 2025, DeepMind introduced Genie 3 as its first world model to run in real time, at 720p and 24 frames per second, with consistency lasting a few minutes. DeepMind presents Genie 4 as a further step, adding native audio generation and live text-based changes. The source also notes that audio generation is not unique to Genie 4: TechCrunch reported that Runway’s GWM Worlds 2 generates audio as well.
A world model predicts what a scene may look like next based on what has already appeared and the user’s input. That differs from a conventional video generator, which produces a clip for later viewing, and from a game engine, which renders a view from a defined environment. DeepMind says Genie 4 has no stored 3D map; it generates the continuing scene as it goes. This design can support flexible interaction, while leaving room for details to change or drift.
““a few minutes of continuous interaction””
— Google DeepMind
Limits and Release Details
DeepMind acknowledges that objects and scene details may shift or drift and describes continuous interaction as lasting a few minutes. Independent testing, published benchmark results and a clear measure of scene consistency are not included in the source. Reports of scenes lasting about two minutes come from TechCrunch and Bloomberg, as cited by heise, rather than from DeepMind’s announcement text.
The timing and terms of broader access remain unknown. DeepMind has not specified a public launch date, Gemini API price, or which early partners will receive access. The source also does not establish whether the demos’ performance or quality will be available to every user, or how the SynthID watermark will be detected in practice.
Next Steps for Access
DeepMind says Google AI Ultra subscribers will be able to use Genie 4 through Project Genie and that early Gemini API partners are also expected to gain access. Both are described as coming soon, but no dates or API pricing have been provided. The next useful milestones will be a dated access announcement, published product terms and testing beyond the company’s selected demonstrations.
Until then, the announced use cases remain preliminary: tester feedback points to film and game planning, while the robotics example is a simulation. Whether developers can rely on Genie 4 for longer sessions or repeatable production workflows will depend on how the model performs under broader testing.
Key Questions
Who can use Google Genie 4 now?
Google DeepMind says access is limited to trusted testers. It has not announced general public availability.
Who is expected to get access next?
DeepMind says Google AI Ultra subscribers through Project Genie and early Gemini API partners are expected to get access. No date or API price has been announced.
What can people build with Genie 4?
The demonstrations show interactive worlds that users can steer and change with text. Tester feedback reported by DeepMind points to film previsualization and game prototypes, but these are early use cases rather than confirmed production capabilities.
Can Genie 4 train robots?
The source describes a robotics simulation involving a humanoid playing checkers, with no physical robot involved. Waymo uses the predecessor Genie 3 in its driving simulator, but the source does not say that Waymo uses Genie 4.
What are Genie 4’s known limits?
DeepMind says scene details can drift and that continuous interaction lasts a few minutes. It has not published benchmark results in the supplied material.
Source: Thorsten Meyer AI
