# Google Genie 4 Explained: DeepMind's World Model Now Makes Sound

- URL: https://jemiboo.com/en/tech/2026-11-02/
- Topics: Tech
- Published: 2026-11-02
- Publisher: Jemiboo Magazine (JEMIBOO)
- Tags: Google Genie 4, Genie 4, world model, Google DeepMind, Project Genie, Gemini, AI world simulation, robot training simulation, SynthID, Google AI Ultra, AI video generation

![World model concept image: a virtual landscape forming inside a glass cube](https://jemiboo.com/img/2026-11-02-tech-cover.jpg)

Google's news last week was not only about Gemini 4 Carbon. A day earlier, DeepMind unveiled Genie 4, a world model that builds an interactive virtual world from a text prompt or a single image, and this time it generates sound on its own as well. It is still at the testing stage, but here is a look at how far it has come.

## Key takeaways (as of November 2, 2026)
- Google DeepMind unveiled Genie 4, a real-time world model, on October 27, 2026, and began offering it as a research preview to a limited group of testers.
- Genie 4 is the first of Google's real-time world models to generate audio such as music, ambient sound and dialogue directly alongside the visuals, and it lets users change the weather or add characters with text prompts.
- Google plans to roll Genie 4 out gradually to Google AI Ultra subscribers through Project Genie and to select partners on the Gemini API and Google AI Studio, but as of November 2, 2026, it has not announced a date.

## 1. What is Google Genie 4?

A world model is an AI model that generates the next scene in real time in response to what the user does. Instead of playing back a pre-rendered video, it creates the view ahead when you move forward with the keyboard and the view to the side when you turn your head. Genie 4 adds three things to this approach. The first is sound. The model itself produces wind, waves, instrumental music and character dialogue, with no separate audio work. Google says Genie 4 is the first of its real-time world models to generate audio directly. The second is editing while the world is running. Type something like “make it rain” and the weather changes, the camera moves, or new people and objects appear. The third is responsive characters. Virtual characters keep a conversation going in response to user input and to what is happening around them.

| Feature | Genie 3 (announced August 2025) | Genie 4 (unveiled October 27, 2026) |
|---|---|---|
| Visuals | Real-time virtual world generation | Real-time generation with improved realism and motion |
| Audio | None | Generates music, ambient sound and dialogue |
| Editing while running | Demonstrated at announcement, but left out of Project Genie | Change weather, camera, characters and objects with text |
| Public access | Opened to U.S. Ultra subscribers via Project Genie in January 2026 | Research preview; general availability date not set |
| Watermark | - | SynthID applied to all video and audio |

## 2. Training robots in a virtual world first

The most notable part of the announcement concerns robots rather than film or games. Google paired Genie 4 with Gemini to carry out multi-step tasks in a virtual environment. Gemini 3.8 Flash breaks a goal into steps and, by watching the generated scenes, checks whether each step has been completed. Gemini 4 Argon passes instructions along as text, and Genie 4 renders the corresponding scene. One example shown was Apollo, a humanoid robot, playing a game of checkers. This was not an experiment with a physical robot but a simulation built in a virtual world. Testing robots in the real world takes space, equipment and time. If robots can first work through a large number of situations in a virtual world, development costs could come down, and that is why the industry is paying close attention to world models.

![A real robot shown side by side with its counterpart in a virtual simulation](https://jemiboo.com/img/2026-11-02-tech-inline.jpg)

## 3. When will it be available, and who can use it?

For now, only a small group of testers selected by Google can use it. Google has said it will extend access to Google AI Ultra subscribers through Project Genie and to early access partners on the Gemini API and Google AI Studio, but it has not given specific dates. Genie 3 followed a similar path. After its announcement in August 2025, it did not open up until January 29, 2026, when it became available as Project Genie to Ultra subscribers in the U.S. aged 18 and over, with each session limited to 60 seconds. Readers outside the U.S. should probably expect a longer wait before they can try Genie 4 themselves.

| Date | Milestone |
|---|---|
| August 2025 | Genie 3 announced (research) |
| January 29, 2026 | Project Genie opens to Google AI Ultra subscribers in the U.S. |
| October 27, 2026 | Genie 4 unveiled; research preview for testers |
| Planned (no date set) | Project Genie Ultra subscribers; Gemini API and AI Studio partners |

## 4. Limitations that remain

Google is upfront about the limitations. When a session runs long or edits are applied repeatedly, the appearance and position of objects can gradually drift. Complex physics, the accuracy of real-time audio, on-screen text, spatial relationships and long-term consistency all remain open research problems. Every video and audio clip Genie 4 generates carries SynthID, Google's watermark for identifying AI-generated content. The move appears to respond to concerns that realistic fake videos could become more common.

This looks like the same pattern Google followed with Gemini 4 Carbon, which it also chose not to release to everyone at once: the more capable the model, the more likely it is opened to a small group first and expanded slowly. How virtual worlds that generate both picture and sound will change game and robot development probably won't become clear until Genie 4 reaches Ultra subscribers.

## Frequently asked questions
### Can I use Google Genie 4 now?
As of November 2, 2026, Genie 4 is available only as a research preview to a limited group of testers selected by Google. Google has said it will expand access to Project Genie Ultra subscribers and API partners, but it has not announced dates or countries.
### What is the difference between a world model and AI video generation?
AI video generation produces a clip of fixed length in one go. A world model generates the next scene in real time each time the user moves or gives a command, so you can explore a virtual space as if you were walking through it.
### Can you tell whether a video was made with Genie 4?
Google says it embeds a SynthID digital watermark in every video and audio clip Genie 4 produces. The watermark is invisible to the eye, but Google's detection tools can confirm whether content was generated by AI.

## Sources
1. Google DeepMind, Genie 4: a new generation of world simulation (October 27, 2026)
2. ZDNet Korea, Google unveils world model Genie 4 to build virtual worlds for training robots (October 28, 2026)
3. Google DeepMind, Genie 3 announcement (August 2025)
4. Google, Project Genie launch for Google AI Ultra subscribers (January 29, 2026)
