jsonscraper

Google unveils Genie 4: Interactive simulations controlled with natural language

The model generates editable scenes and sound in real time. Testing has begun with a limited group, while robotics and development are cited as potential applications.

Editorial Policy Report an error

Genie 4 simulations can be changed while they are running: use a text prompt to move the camera, change the weather, or add a character. Google announced Genie 4 on October 27, 2026 as a research world model for interactive scenes with generated sound.

Article: Google unveils Genie 4: Interactive simulations controlled with natural language
Summary: On October 27, 2026, Google DeepMind introduced Genie 4, a research world model for interactive simulations. Users can change what happens with text prompts, while the system generates audio on the fly. Its practical value will depend on future access to the tools and the ability to test simulations against specific tasks.
Image role: cover — the central idea
Section: 
Visual subjec
© jsonscraper · AI-generated illustration

Google said it had begun testing with a limited group of trusted users. The company says access for Google AI Ultra subscribers through Project Genie and for early-access partners through the Gemini API and AI Studio will follow later. These are announced future access channels, not an API already available to the general public.

A scene you can control

According to Google, Genie 4 uses capabilities from the Gemini family: Gemini Omni helps model the appearance and behavior of the world based on the real world, while other Gemini models provide the reasoning needed to keep the simulation coherent and continue it. Users can direct the scene in natural language; the system generates ambient sounds, music, and interactive dialogue on the fly.

This approach shifts generation from making a standalone video to controlling an evolving environment. For creative work, it could speed up the exploration of scene variations. Google cites robot and agent training, as well as film, journalism, games, and production, as other potential applications.

What developers should evaluate

Teams assessing the model for a specific task should test it using their own scenarios. For robot training, important factors include the world's stability under repeated commands, whether the simulation matches the required physics, and whether experimental conditions are reproducible. For games and media scenes, key considerations include control over actions, audio-visual consistency, and the ability to control generated content.

Google's announcement provides no numerical results for these criteria and does not specify API access terms, limits, or pricing. What has been confirmed is the start of limited testing and the system's stated capabilities; assessing Genie 4 for a particular workflow will require access and reproducible tests.

Genie 4 is worth watching as a model for interactive simulations that can be controlled with prompts. Its practical value for developers will depend on testing availability and how reliably the system supports the scenarios they need.

People

No people listed for this article yet.

Keep readingFour Gemini Agents Beat Minecraft in 27 Hours, Writing 2,783 Programs
Read the next article

Turn what you read into a working integration

Explore jsonscraper's social-data APIs, test requests and build your next workflow.

Explore APIs