Google DeepMind · 2026-10-27 · major
Genie 4 — DeepMind's world model adds native audio and live text steering
Google DeepMind announced Genie 4, a real-time interactive world model with native audio, simulated physics and natural-language steering. Genie 4 is a research preview for trusted testers, and broader access has a waitlist.
A world model you can steer with words while it runs, now with sound, speaking characters and simulated physics.
Quick facts
| Maker | Google DeepMind |
|---|---|
| Type | Real-time interactive world model |
| Built on | Gemini Omni + Gemini 3.8 Flash as a guiding agent |
| What's new | Native audio, language steering, simulated physics |
| Availability | Research preview for trusted testers; waitlist |
| Coming next | Project Genie (AI Ultra), Gemini API, AI Studio |
| Watermark | SynthID on all video and audio |
What is it?
Genie 4 is Google DeepMind's new world model: it generates an interactive environment in real time that a user can explore and change while it runs. DeepMind calls it natively multimodal, because ambient sound, music and dialogue are generated together with the picture, and characters can talk with the user and with each other.
How does it work?
Gemini Omni anchors the look and behavior of each simulation in real-world appearance, while other Gemini models handle reasoning and continuity. Gemini 3.8 Flash acts as a guiding agent that plans steps and checks progress during a session. Users type commands to move the camera, change the weather or add people and objects, and the model simulates effects such as collisions, flowing water and aerodynamics.
Why does it matter?
World models are a training ground for agents and robots, and DeepMind describes early work pairing Genie 4 with Gemini Robotics. Adding sound, dialogue and language control moves these simulations closer to usable environments for games, film previs and robot training, though DeepMind lists drift over long sessions and short session length as open problems.
Who is it for?
world-model researchers, robotics teams, game and media creators
Frequently asked questions
- Can I use Genie 4 today?
- Genie 4 is a research preview limited to trusted testers. Google DeepMind says access is coming soon to Google AI Ultra subscribers through Project Genie, and to early-access partners in the Gemini API and Google AI Studio. Anyone else can join a waitlist linked from the announcement page.
- What are the known limitations of Genie 4?
- Google DeepMind says Genie 4 sessions are limited to a few minutes, and visual details and object permanence can drift over longer sessions or after many edits. Multi-agent interaction, accurate simulation of real places, legible text unless it is in the prompt, and real-time audio accuracy are listed as areas still under research.
- How does Google DeepMind keep Genie 4 output safe?
- Google DeepMind ran internal safety evaluations before the Genie 4 preview, applies safety guardrails consistent with its other models, and limits the first rollout to trusted testers. All video and audio that Genie 4 generates carries a SynthID watermark so it can be identified as AI-generated.
- What can Genie 4 be used for besides games?
- Google DeepMind presents robotics as a key future use for Genie 4 and describes early research pairing the world model with Gemini Robotics. A simulated world with physics, characters and sound gives robots and AI agents a place to practise tasks before they act in the real world.
Try it
Join the Genie 4 waitlist from the DeepMind announcement page.