Genie 4: DeepMind's Interactive World Model Now Makes Sound Too
Google DeepMind's Genie 4, announced last week, generates playable worlds with audio and dialogue in real time. It is a research preview with a waitlist.
Last week, Google DeepMind introduced Genie 4, a “world model.” In plain terms, it is an AI that generates a small, interactive world on the fly, which you can steer as it runs. DeepMind says it is the first real-time world model that also produces sound natively: ambient noise, music and dialogue, not only pictures.
You can change the scene with ordinary sentences while it runs, for example altering the weather, moving the camera or adding a person or object. The model simulates parts of physics such as collisions and airflow, characters move more naturally, and they can talk and react to what you do. Gemini models help keep appearance and behaviour consistent, and in a separate experiment Gemini gave a simulated humanoid step-by-step captions to play checkers. Early testers are exploring previsualisation (sketching a film scene before shooting), recreating historical scenes from archival photos, quick game prototypes and business scenario generation.
Access is limited. Genie 4 is a research preview for trusted testers, with plans for Google AI Ultra subscribers in Project Genie and for early partners in the Gemini API and Google AI Studio. A waitlist is open. DeepMind says everything it generates carries a SynthID watermark, an invisible marker that signals AI origin.
What’s behind it: Why would a company that builds chatbots care about game-like worlds? Because a world that responds to actions is a practice field. Robots and software agents can rehearse in a simulation before touching the real world, and the model has to learn something about how things behave to keep the simulation consistent. It is also a step toward more general AI. DeepMind is open about the limits: details and object permanence can drift in longer sessions or after many edits, and complex physical interactions, accurate audio, readable text, spatial accuracy and long-term consistency are all still being worked on.
What this means for you: If you are curious, join the waitlist and watch the demos, but do not expect to use it this week. Creators and teachers may eventually find it useful for quick visual drafts. A fair expectation for now: impressive demos, rough edges over time. When you see AI-generated video or audio, features like SynthID are a good reminder to look for provenance labels.
Sources
Source: https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
Gemini 4 Carbon: Google's Strongest Model Goes to Cyber Defenders First
Google announced Gemini 4 Carbon last week with top scores in security tasks, limited access and mixed independent results on accuracy.
