Google DeepMind has released details about Genie 4, the newest generation of its Genie world-model series. Access is being offered initially as a research preview to selected trusted testers, with plans to make the model available later to Google AI Ultra subscribers.
The Genie models aim to turn images or 2D drawings into interactive environments — for example, converting a photo into an explorable scene or transforming a sketch into a videogame stage. DeepMind highlights four key improvements in Genie 4 compared with earlier versions.
Key features
-
Real-time modification
- Users can alter a simulation in real time using natural language: change the weather, move the camera, or introduce new people or objects. The system adapts prompts contextually to fit the surrounding scene.
-
Physics and human-action simulation
- Genie 4 simulates real-world physics scenarios such as aerodynamics and collisions. The model also advances how it simulates people: characters move more naturally, respond organically to their environments, and can be interactively steered in real time.
-
Sound modeling
- This release is DeepMind’s first real-time world model to generate audio natively. It also introduces interactive dialogue that adapts on the fly to the current situation.
-
Simulating complex events with continual calibration
- Genie 4 embeds the Gemini 3.8 Flash model so that the fast Flash model acts as a guiding agent in the loop. Gemini continuously observes the scene, decomposes goals into sequences of steps, and reviews incoming frames to track progress and verify that each step succeeded. As the world shifts or the user intervenes, Gemini adapts its plan in the moment, enabling reliable causality, procedural behaviors, and multi-step reasoning.
Why it matters
DeepMind acknowledges that the technology is not yet polished: demonstrations can include robotic-sounding voices and dreamlike inconsistencies in visualized environments. Nonetheless, Genie 4 points toward a future in which custom, machine-generated worlds could create new forms of entertainment and potentially new kinds of data generation.
Accompanying piece: Tech Tales — City Race
Alongside the announcement, DeepMind shared a short fictional piece titled "City Race." The story describes a 2040s-era game in which children plan, build, and manage cities on a novel planet using a slice of resources from a Dyson Swarm to power a highly complex simulation. It depicts hundreds of thousands of simulated minds living through different urban outcomes and frames the game as a way for players to learn godlike city-building skills before adulthood.
Genie 4 represents a significant step in real-time, multimodal world modeling, but the company’s own notes about current limitations indicate that further refinement will be needed before the technology is ready for broad deployment.
