Imagine typing “a crowded rural train station in 1980s Japan” and then walking around inside it, hearing the platform noise, changing the weather with a sentence, and watching passengers react. That is the promise of Genie 4, the new world model Google DeepMind announced on October 27, 2026. It is still a research preview, but it shows how quickly AI is moving from generating pictures and clips to generating whole interactive environments.
What a “world model” does
A world model is an AI system that learns how environments look and behave, then predicts what happens next as you act in them. Instead of rendering a scene from hand-built 3D assets, as a game engine does, it generates each frame on the fly from what it has learned. DeepMind frames this as central to its long-term goals, writing in its launch post that “understanding and simulating the world is a crucial aspect of building AGI.”
Four upgrades in Genie 4
DeepMind groups the advances into four capabilities:
- Real-time shaping with language. Users can change the weather, move the camera, or introduce new people or objects by typing, and Gemini adapts the prompt to fit the scene. Between inputs, Gemini also monitors the simulation to keep it progressing coherently.
- More realistic physics and people. The company says Genie 4 simulates aerodynamics and collisions and, most notably, characters that “move naturally, respond organically to their environments, and can be interactively steered in real-time.”
- Native audio. Genie 4 is DeepMind’s first real-time world model to generate sound natively, from wind and waves to interactive dialogue in which characters converse and react to the user.
- Planning with Gemini in the loop. Gemini 3.8 Flash acts as a guiding agent: it watches the frames, breaks goals into steps, and checks whether each step succeeded, adapting the plan when the world or the user changes.
Under the hood, DeepMind says Gemini Omni “anchors the simulations in real-world appearance and behavior,” while other Gemini models supply the reasoning needed for continuity.
Why it matters beyond entertainment
Games and films are the obvious use cases, and DeepMind reports that trusted testers in creative industries are using Genie 4 as a “draft mode” for pre-visualization, running multiple takes of a scene and adjusting camera, lighting and dialogue on the fly. Game developers are testing it to move from an idea to an explorable prototype before writing engine code.
The bigger prize may be training AI agents and robots. Real-world robot data is slow and expensive to collect, and some scenarios are impractical to stage. A model that can produce endless, varied, realistic environments offers a safer place to practice. DeepMind showed Gemini 4 Argon breaking a task into steps that Genie 4 renders in real time, including a simulation of Apptronik’s Apollo humanoid playing checkers. The company calls this “an infinite playground for robotics simulations” on its model page.
The honest limitations
DeepMind is unusually direct about what Genie 4 cannot yet do. According to its model page:
- Sessions support “a few minutes of continuous interaction,” not extended use.
- Visual details and object permanence can drift, especially in longer sessions or after many prompt edits.
- Real-world places are not reproduced with perfect accuracy.
- Legible text usually appears only when it is in the prompt, and real-time audio accuracy is still being improved.
On safety, the company says Genie 4 went through internal evaluations before the preview, includes guardrails consistent with its other models, and marks all generated video and audio with SynthID, its imperceptible watermarking technology. An independent weekly review by The Model Press noted the same limits and that the model is restricted to trusted testers for now.
Availability
Genie 4 is rolling out as a research preview to trusted testers. DeepMind says it is “coming soon” to Google AI Ultra subscribers in Project Genie and to early-access partners of the Gemini API and Google AI Studio, with a public waitlist open.
Practical takeaways
- Creative teams should watch world models as a fast storyboarding and pre-visualization tool, not yet as a replacement for production pipelines.
- Robotics and simulation teams can start thinking about how generated environments could supplement, not replace, physics-accurate simulators and real-world testing.
- Everyone should expect more synthetic video and audio in circulation, which makes provenance tools like watermarking increasingly important.
The bottom line
Genie 4 is a preview of a new kind of software: worlds that are generated rather than built. Its few-minute sessions and drifting details show how early this is. But combining photorealism, sound, speech and a reasoning model that keeps the story on track is a meaningful step, and one that could reshape how games are prototyped, films are planned and robots learn.