By Hamza Faisal

Google DeepMind has unveiled Genie 4, a new world model that turns a text or image prompt into an explorable, photorealistic virtual world in real time. Announced on October 27, 2026, it is the latest step in DeepMind's push toward AI systems that do not just describe the world, but simulate it — a capability the lab calls crucial to building artificial general intelligence.

Unlike a video generator, which produces a clip you watch, a world model produces a place you move through. Type a sentence and Genie 4 builds a scene frame by frame as you navigate it with keyboard controls, responding to what you do rather than playing back something pre-rendered. DeepMind says the model delivers enhanced photorealism, more fluid movement and stronger physics, including better collisions and aerodynamics, than its predecessor Genie 3.

Worlds that talk back

The headline change is sound. Genie 4 is DeepMind's first real-time world model with native audio: music, ambient noises and interactive dialogue are generated simultaneously with the world itself, not layered on afterwards. Characters in a scene can converse and react in real time, so a world stops being something you watch and becomes something you can talk to.

That matters for more than immersion. DeepMind positions the model as a tool for building training environments for robots and AI agents, as well as new kinds of interactive experiences. A simulated world that looks, sounds and behaves consistently is far more useful for training than a silent one — and the lab says every sound in its demonstration clips was produced directly by Genie 4.

Steer it live, for about two minutes

Genie 4 can also be steered mid-simulation with plain-language prompts. Change the weather, move the camera, or add a person or an object while the scene keeps running, and the model adapts the prompt to fit the world around it. Under the hood, Gemini Omni anchors the simulation in real-world appearance and behaviour, while Gemini 3.8 Flash acts as an in-loop guiding agent, checking frames and breaking high-level instructions into steps to keep the scene coherent.

Scenes now last about two minutes, double the 60-second limit of Genie 3. Two minutes is still short, but it is long enough for a scene to develop: a character can cross a room, weather can turn, a conversation can go somewhere. It is also where the model's limits show — DeepMind is candid that visual details and object permanence can drift over longer runs, especially after multiple complex prompt edits.

A research preview, not a finished product

For now, Genie 4 is a research preview for trusted testers, including filmmakers, game developers and robotics teams. Wider access will follow for Google AI Ultra subscribers through Project Genie and for early-access partners of the Gemini API and Google AI Studio, with a waitlist open for everyone else. DeepMind has not announced a firm timeline for general availability.

The launch lands in an increasingly crowded field, with rivals including Nvidia's Cosmos and Runway's world models chasing the same idea: AI that simulates reality well enough to train in and play in. Genie 4's answer is completeness — a world with sound, steering and staying power, even if only for two minutes at a time. For creators, it is a glimpse of a workflow where describing a scene and directing it become the same act.