Models · 2026-10-28

Google unveils Genie 4, an AI that builds interactive worlds in real time

Google announced Genie 4, a neural network that creates realistic simulations on the fly, responds to user commands, and generates physics and sound.

✨ What the model is

Google announced Genie 4, a multimodal neural network that creates interactive simulations and reacts to a person's commands instantly. The company calls it a step toward universal artificial intelligence. The model generates not only images but also object behavior: physics, collisions, aerodynamics, and natural movement of people in the frame.

The model also creates sound itself. Genie 4 adds background noise, music, and human speech on its own. Characters can hold dialogues with the user in real time.

📌 How to control it

You can control the world with standard keyboard keys or text prompts. You can change the weather, move the camera, or add new characters during a simulation. The system adjusts these changes to the context of the scene.

The Gemini 3.8 Flash model is responsible for the logic and continuity of events. It analyzes frames, breaks large goals into smaller steps, and adjusts the plan if the user intervenes. In parallel, Google is testing integration with Gemini 4 Argon to build training environments for robots.

🔹 Where it is being tested and its limits

The tool is being tested in several industries. Game developers turn ideas into interactive prototypes without writing code. Directors quickly try different angles and lighting. Photojournalists turn flat archival photos into dynamic spatial locations.

Google acknowledges its limitations. During long sessions or after many edits, objects in the simulation can disappear or deform, and text rendering needs improvement. All generated content is automatically marked with the SynthID digital watermark to help distinguish it from real footage.

🔸 When access opens

Genie 4 is currently available only as a closed research preview for trusted partners. Later, access will go to Google AI Ultra subscribers through Project Genie, as well as to developers via the Gemini API and Google AI Studio.

Sources: 🔗 DOU/dev.ua