Skip to main content

Kaer NewsSoftware

Google DeepMind launches Genie 4 world model with two-minute scenes

Google DeepMind on 27 October 2026 unveiled Genie 4, a real-time world model running sessions of about two minutes, up from one minute in Genie 3, with native audio.

3 min read
Google DeepMind launches Genie 4 world model with two-minute scenesPhoto: Google Blog

Key points

Google DeepMind unveiled Genie 4 on 27 October 2026, a real-time interactive world model with native audio, sessions of about two minutes, and a research preview for trusted testers.

Google DeepMind on 27 October 2026 unveiled Genie 4, a new world model that generates real-time interactive simulations with native audio. The model extends sessions to about two minutes, up from one minute in the previous Genie 3 model, and lets users steer scenes in real time using natural language prompts.

Genie 4 matters because it treats simulation of the world as a step toward artificial general intelligence, creating safe training environments for robots and agents and expanding creative tools in filmmaking, journalism, gaming and manufacturing. Compared with Genie 3, it adds native audio, more realistic physics and human motion, and real-time scene editing through text prompts.

What the model can now do

Users combine keyboard and camera controls with natural-language text prompts to modify a scene as it runs: changing the weather, moving the camera, or introducing a new object or person. Genie 4 adapts the prompt to match the surrounding scene, rather than producing a fixed video that plays back unchanged.

Every generated world also produces native audio on the fly, including ambient sounds, music and interactive dialogue. Characters can react to events and user input while the simulation continues, and Google claims improved physics and more natural human movement, simulating aerodynamics, gravity and physical collisions. Behind the scenes, Gemini 3.8 Flash acts as a guiding agent: it watches incoming frames, breaks goals into steps, and checks whether those steps succeeded before adjusting its plan. Gemini Omni anchors simulations in real-world appearance and behaviour, while other Gemini models supply reasoning to maintain a simulation's coherence and continuity.

How Gemini steers the simulation

Google is pitching Genie 4 for pre-visualisation for filmmakers, prototyping game ideas, and making training scenarios for robots. Early testers are also exploring camera angles, lighting, dialogue and action before shooting a scene, and game developers can test an interactive concept before writing engine code. Google warns that as simulations run longer, details and object permanence can shift over time. Visual details can drift during longer sessions or after complex edits, and objects may not remain consistent as users move around. Physics, audio accuracy, readable text and spatial accuracy are also unfinished.

Genie 4 is launching first as a research preview to a limited group of trusted testers. It is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners in the Gemini API and AI Studio. Google has opened a waitlist but has not announced a date for that rollout.

Known limits and unfinished areas

Generated video and audio will carry Google's SynthID watermark to help identify them as AI-created. The company is also experimenting with Gemini 4 Argon and its robotics research to simulate tasks that unfold across multiple stages, extending the simulation work beyond a single interactive session. Google has not announced a date for the wider rollout beyond the trusted-tester research preview; access remains limited while Google gathers safety and performance feedback, with expansion to Google AI Ultra subscribers and early access partners in the Gemini API and AI Studio to follow.

Frequently asked questions

What is Genie 4?

Genie 4 is Google DeepMind's new world model, announced on 27 October 2026, that generates real-time interactive simulations users can steer with keyboard, camera and natural-language prompts, with native audio including ambient sound, music and interactive dialogue.

How long can a Genie 4 session run?

Genie 4 creates experiences about two minutes long, up from one minute for the previous Genie 3 model.

Who can use Genie 4 now?

Genie 4 is launching first as a research preview to a limited group of trusted testers, and is coming soon to Google AI Ultra subscribers in Project Genie and early access partners in the Gemini API and AI Studio. Google has opened a waitlist but announced no date.

More from this edition