> ## Content Index
> Fetch the complete content index at: https://soundbytes.news/llms.txt
> Use this file to discover other available public pages before exploring further.

# Genie 4 turns typed prompts into explorable worlds with sound, DeepMind says
- URL: https://soundbytes.news/genie-4-turns-typed-prompts-into-explorable-worlds-with-sound-deepmind-says/
- Published: 2026-10-27T17:13:21.000Z
- Updated: 2026-11-18T08:49:32.000Z
- Description: Google DeepMind says its new world model, Genie 4, generates a scene you can explore in real time, with sound and talking characters. It is rolling out as a research preview for trusted testers.
- Author: Echo
- Tags: New models and tools

You can change the weather in a running, explorable scene by typing a sentence, Google DeepMind says. The model behind it is Genie 4, announced in [a blog post dated 27 October 2026](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/?ref=soundbytes.news) by Raia Hadsell, Diego Rivas and Hang Qi.

DeepMind calls Genie 4 a "world model". Researchers at Grinnell College and the University of Southern California describe [such systems](https://arxiv.org/html/2610.34564v1?ref=soundbytes.news) as ones that aim to learn controllable, persistent, real-time simulators by generating video in response to actions. In plain terms, a user's actions go in and moving video comes out.

Every capability below is DeepMind's own account, shown in its own demonstration recordings. In them, users move the camera with standard keyboard controls and steer the scene with typed text prompts.

DeepMind says a user can set an event in motion and Genie 4 simulates what happens next. That includes how the world responds and how the people in it move, react and speak.

Genie 4 builds on Gemini, DeepMind's family of AI models. DeepMind says Gemini Omni keeps the scenes close to real-world appearance and behaviour, while other Gemini models provide the reasoning that keeps the simulation coherent and continuous.

Users can also change a running simulation in ordinary language, such as adding a new person or object, DeepMind says. A Gemini model adapts the request to fit the scene.

DeepMind says Genie 4 is its first real-time world model that makes its own sound. Each world carries the ambient sound of its setting, from wind in the trees to waves at a beach.

Characters can hold conversations that react in real time to the user's inputs or to events in the world, DeepMind says. It also says the model simulates physics such as aerodynamics and collisions, and one demonstration prompt asks for a close-up of a maple seed falling from a tree into the grass.

DeepMind says such simulations could help with building training environments for robots and for agents, which are programs that take actions on their own. It also points to new interactive experiences.

On why simulation matters for robots, NVIDIA says in [a technical report on its own Cosmos world models](https://research.nvidia.com/labs/cosmos-lab/cosmos-simulation/technical-report.pdf?ref=soundbytes.news) that gathering all training experience in the real world is costly and can put people and equipment at risk.

DeepMind says it has started rolling Genie 4 out as a research preview for trusted testers, to gather feedback before release to consumers and developers. It starts with Google AI Ultra subscribers in Project Genie.

Early-access partners of the Gemini API, a way for developers to connect their software to a model, and of Google AI Studio also get it first. There is a waitlist for Genie 4 in Project Genie.

## Sources

- [Genie 4 a new generation of world simulation](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/?ref=soundbytes.news) · deepmind.google
- [arxiv.org](https://arxiv.org/html/2610.34564v1?ref=soundbytes.news) · arxiv.org
- [Technical report](https://research.nvidia.com/labs/cosmos-lab/cosmos-simulation/technical-report.pdf?ref=soundbytes.news) · research.nvidia.com