TL;DR
Get everyday essentials 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 demonstrated Genie 4 on October 27, 2026, showing an AI model that generates interactive scenes from prompts and responds to movement and text instructions. The company says it can create matching audio and more realistic motion, but published no benchmark figures; access is limited to trusted testers, and scenes may drift during use.
Google DeepMind demonstrated Genie 4 on October 27, describing it as a real-time AI world model that creates scenes from text prompts and lets users steer them with controls and further instructions. The announcement points toward interactive, sound-producing environments for uses such as game prototyping, but the system is not publicly available, and DeepMind says scenes can shift or drift during extended interaction.
In the company’s demonstrations, a user can enter a scene, move through it with keyboard or mouse controls, and type changes while it runs, such as making the weather rainy or adding an object. DeepMind says Genie 4 generates images and audio together, including ambient sound, music and dialogue. These capabilities are presented in company-selected clips; DeepMind has not published benchmark results that independently quantify the model’s performance.
DeepMind says Gemini 3.8 Flash works alongside the model to interpret goals, break them into steps and monitor generated frames. The company also describes more realistic people and physical interactions, including collisions and aerodynamics. Its robotics example showed a simulated Apptronik Apollo humanoid playing checkers; according to the source report, no physical robot was involved.
Access is currently restricted to trusted testers. Google said access is planned for Google AI Ultra subscribers through Project Genie and for early Gemini API partners, but it has not given a release date or API pricing. The source material identifies Google AI Ultra as a $200-per-month tier at the time of publication. DeepMind says generated video and audio carry an invisible SynthID watermark.
Interactive world models · October 27, 2026
How Google’s Genie 4 Turns Prompts Into Worlds You Can Explore
Google DeepMind says Genie 4 can turn a text prompt into a live scene that responds to movement and new instructions. Its demos point toward worlds with generated sound, while access, independent performance data and long-term scene stability remain open questions.
The central idea
“A few minutes of continuous interaction”Google DeepMind’s description
At a glance: Genie 4 is a demonstrated research and product development, currently limited to trusted testers.
01 / The development
Interactive worlds beyond video
DeepMind presents Genie 4 as a real-time world model: it generates a scene from a prompt, then predicts what happens next as a user moves or asks for changes.
Start with a prompt
Describe a scene and enter a generated environment. In demonstrations, the user explores it with keyboard or mouse controls.
Change it as it runs
Type instructions during interaction, such as making the weather rainy or adding an object to the scene.
Generate sound, too
DeepMind says images and audio are generated together, including ambient sound, music and dialogue.
Prompt
Describe the world you want to explore.
Generate
The model creates the scene and its next frames.
Interact
Move through the scene with controls.
Instruct
Ask for changes while the world is running.
02 / Evidence and limits
Promising demonstrations, unanswered tests
DeepMind says Gemini 3.8 Flash works alongside Genie 4 to interpret goals, break them into steps and monitor generated frames. The company also describes more realistic people and physical interactions, including collisions and aerodynamics.
Its robotics example showed a simulated Apptronik Apollo humanoid playing checkers. The source report says no physical robot took part in that demonstration.
DeepMind says generated video and audio carry an invisible SynthID watermark. The supplied material does not establish how reliably the mark can be detected or retained after editing.
03 / A moving frontier
Genie’s path toward real time
The announcement builds on earlier world-model work. These milestones describe the product’s development as reported in the source material; they are not independent performance comparisons.
Genie
Generated simple 2D worlds from video training data.
Genie 2
Generated 3D environments from a single image and maintained them for up to a minute; real-time use required reduced quality.
Genie 3
Described by DeepMind as its first real-time world model, at 720p and 24 frames per second, with consistency for a few minutes.
Genie 4
Presents more realistic scenes, native audio and text-based changes during interaction.
04 / Access and reliability
What is known—and what is still open
Restricted testing
Access is limited to trusted testers. DeepMind has described planned access through Project Genie for Google AI Ultra subscribers and for early Gemini API partners.
Google AI Ultra: $200/month at publicationTiming and API terms
No general release date or Gemini API pricing has been provided. The source material also does not establish whether all planned features will reach subscribers and partners.
Questions wider testing needs to answer
- Consistency: How long do details stay coherent, and under what conditions?
- Duration: DeepMind says “a few minutes”; outside reports cited in the source describe up to about two minutes, a figure not in its announcement.
- Usefulness: Do curated demos translate into routine game, film or robotics workflows?
- Provenance: How reliably can SynthID be detected after editing?
05 / Traceability
From idea to an explorable draft
The proposed workflow links a creative goal to a generated environment, live interaction and an early concept to evaluate.
Interactive Worlds Beyond Video
Genie 4’s stated distinction from a conventional video generator is that a user can influence what happens while a scene is being created. That makes the announcement relevant to developers considering game concepts, virtual environments and film previsualization, where interactive drafts could help teams explore ideas before building finished assets. DeepMind says testers have reported using the system for film previsualization and game prototypes; those are early user reports, not evidence of established production use.
The approach also differs from a standard game engine. According to the source material’s description, Genie 4 does not rely on a stored 3D map of a scene; it predicts what comes next from what it has generated and the user’s input. That may allow flexible scene creation, while making persistent details harder to maintain. For readers, the combination of generated visuals, sound and live interaction also raises practical questions about how synthetic media is identified. DeepMind’s watermark claim addresses provenance, but the material does not establish how reliably the mark can be detected or retained after editing.
Popular right now
Dyson Cyclone V10 Origin Cordless Vacuum
Ninja CREAMi 7-in-1 Ice Cream Maker, NC301
Philips 5500 Series Fully Automatic Espresso Machine with LatteGo
As an affiliate, we earn on qualifying purchases.
Genie’s Move Toward Real Time
Genie 4 follows earlier versions of DeepMind’s world-model work. The source material says the first Genie, introduced in February 2024, generated simple 2D worlds from video training data. Genie 2, announced in December 2024, generated 3D environments from a single image and maintained them for up to a minute, although real-time operation required a reduced-quality version.
In August 2025, Genie 3 was described by DeepMind as its first real-time world model, running at 720p and 24 frames per second and maintaining consistency for a few minutes. The current announcement presents Genie 4 as a step toward more realistic scenes, native audio and text-based changes during interaction. These are descriptions of the product’s development, not a substitute for independent comparisons of performance.
There is also a distinction between the company’s own description and outside reporting on duration. DeepMind says Genie 4 supports “a few minutes of continuous interaction” and acknowledges that details can drift. The source material cites TechCrunch and Bloomberg, as reported by heise, for scenes lasting up to about two minutes; it says that specific figure does not appear in DeepMind’s announcement.
““a few minutes of continuous interaction””
— Google DeepMind
Access, Stability and Independent Tests
Several points remain open. DeepMind has not announced when general access will begin, how much Gemini API access will cost, or whether all planned features will be available to subscribers and partners. It has also not released benchmark data in the supplied material, so the demos do not establish comparative performance against other systems.
Consistency is another stated limitation: objects and details may drift, and the company describes interactions lasting a few minutes. The exact session duration and the conditions under which scenes remain stable are not fully specified. It is also unclear how the system performs outside selected demonstrations or whether the reported early uses will translate into routine production workflows. The SynthID watermark is a company-reported feature; details about its detection and durability are not provided here.
Broader Testing and Release Details
The next identifiable step is access for Google AI Ultra subscribers through Project Genie and early Gemini API partners, as described by DeepMind. The company has not supplied a launch date or pricing for API access, so the timing and scope of that rollout remain unknown. Wider testing should show whether users can maintain coherent scenes for the stated interaction window and how useful the controls are beyond curated examples.
For now, Genie 4 should be understood as a demonstrated research and product development, not a generally available creation tool. Further announcements, access terms, independent testing and evidence from real development workflows will clarify whether its interactive worlds can be used reliably for games, film planning or robotics research.
Key Questions
What is Google DeepMind’s Genie 4?
Genie 4 is a world model that DeepMind says can generate an interactive scene from a text prompt, then respond to user movement and further instructions.
Can the public try Genie 4 now?
No. The source material says access is limited to trusted testers. DeepMind has described planned access for Google AI Ultra subscribers through Project Genie and early Gemini API partners, but has not announced a date.
Does Genie 4 generate sound as well as images?
DeepMind says the model generates audio with the scene, including ambient sounds, music and dialogue. The claim comes from the company’s announcement and demonstrations.
What are Genie 4’s known limitations?
DeepMind says details and objects can shift or drift over time and describes continuous interaction lasting a few minutes. It has not published benchmark results in the supplied material.
Is Genie 4 being used to train real robots?
The cited robotics example is a simulation of an Apptronik Apollo humanoid playing checkers. The source material says no physical robot was involved in that demonstration.
Source: Thorsten Meyer AI
