Spatial Insider
← RoboticsAI World Models

Google Unveils Genie 4, Its Most Realistic AI World Model Yet

Genie 4 combines Gemini’s reasoning with more realistic physics, natural movement, and generated sound and dialogue. Trusted testers are getting access first.

· 2 min read

ShareXFacebookLinkedIn
An AI-generated saxophone player stands beside a glowing sphere and a lake in Google’s Genie 4 demo.
A saxophone player in Google’s Genie 4 demo. The model generates music and other sounds along with the scene.Image: Google DeepMind · AI-generated scene

Google DeepMind has unveiled Genie 4 (opens in a new tab), which it calls its most realistic and interactive world model yet, and begun rolling it out to trusted testers.

Set an event in motion, and Genie 4 simulates what follows. Gemini helps plan the steps, while the model generates how people move, objects interact, and the scene sounds.

Change the scene as it runs

Google says Genie 4 handles collisions and other physical interactions more realistically, while people move and react more naturally. Use a keyboard and mouse to move around, then type instructions to change the weather, direct a character, or start an action.

In Sundar Pichai’s demo (opens in a new tab), a motorcycle waits at a red light beside the ocean. The instruction “traffic light green” changes the signal. The engine revs, and the bike pulls away down the coast.

The 50-second demo shows typed instructions and steering controls changing the ride. Google says the model generates the sound, too.Video: Google DeepMind

Gemini plans the next steps

Gemini helps decide how the scene should respond. Google says Gemini 3.8 Flash watches the incoming frames, breaks a request into steps, and checks the results as the world changes.

Google says filmmakers are testing Genie 4 to try camera angles, lighting, dialogue, and action before a shoot. In another demo, Gemini plans steps for a simulated Apollo robot playing checkers.

What to expect

Google’s model page (opens in a new tab) lists a few minutes of continuous interaction. Details can change or disappear over time, especially after several complex edits. Physics, sound, readable text, and accurate real-world locations still need work.

Who can try it

Genie 4 is rolling out to trusted testers as a research preview. Google AI Ultra subscribers will get access through Project Genie later. Early-access partners will also get it through Google AI Studio and the Gemini API, Google’s developer interface.

Google has not announced an exact rollout date or separate Genie 4 pricing. You can join the Project Genie waitlist (opens in a new tab) now.

More from Robotics.

Browse all stories
A captured living room becomes a gingerbread-style 3D space in an Echo-3 demonstration.

AI 3D Capture · 2 min read

SpAItial’s Echo-3 Turns iPhone Scans Into 3D Worlds You Can Restyle

The new capture app rebuilds spaces at real-world scale and fills in missed areas. A launch demo turns a family home into a storybook.

Bambu Lab’s dark S1 handheld 3D scanner, showing its front sensor window, buttons, and part of its grip in official teaser artwork

3D scanning · 1 min read

Bambu Lab Announces the S1, Its First 3D Scanner

The handheld scanner is due in November, giving makers a new way to bring physical objects into their designs.

Apple promotional image of a Vision Pro wearer on a sofa surrounded by an underwater scene with sharks and a diver

Apple Immersive · 2 min read

Apple Plans Immersive Video Creator Event in Cupertino on November 11

Creators can preview live production in Apple Immersive, explore new workflows with Apple experts, and join in person or online.

Preparing the Spatial Insider studio