Google DeepMind Genie 4 Guide

Genie 4 is Google DeepMind’s newest world model, announced on October 27, 2026, as a research preview. A world model generates an interactive environment you can move through and act in, in real time, instead of a fixed video.

A video model shows you a world. A world model lets you step into it.

• Native audio: sound and dialogue generated by the model itself, a first for DeepMind’s real-time world models
• Better simulation of people and characters
• More realistic physics, such as collisions and aerodynamics
• Consistent scenes for up to about two minutes, double Genie 3
• Real-time steering of events as you explore
• A Gemini model in the loop to plan multistep events

• A research preview for trusted testers
• Rolling out first to Google AI Ultra subscribers through Project Genie
• Early access partners of the Gemini API and Google AI Studio

It is not generally available. Check Google DeepMind’s Genie page for access updates.

Video models

Veo, Kling, and Seedance make fixed clips you watch

World models

Genie makes environments that respond to your actions as you go

Games

Rapid prototypes of levels and worlds

Education

Explorable historical or scientific scenes

Film and design

Exploring locations and ideas before production

Robotics

Simulated environments to train and test robots

AI research

Training agents in rich, varied worlds

A misty medieval harbour at dawn, wooden boats creaking, gulls calling, market stalls opening along the dock. I walk along the pier as fishermen unload their catch.

• Describe the place, time, and mood
• Include sounds you expect to hear
• Say what you want to do in the world

• An early research preview, not a finished product
• Longer sessions can show visual inconsistencies
• Limited access for now
• Real-time generation needs heavy computing

• Do not create realistic worlds that mislead people about real events
• Respect copyrighted characters and places
• Follow Google’s generative AI policies

• Wider availability beyond trusted testers
• Longer, more consistent sessions
• Tools for game and app makers
• Robotics results from simulation training