Imagine you type "rainforest at night" and moments later you are standing in it. You swing the camera, make it rain, and somewhere a frog croaks.
This is not a film scene someone rendered in advance. The AI generates it while you watch.
What Genie 4 is
Google DeepMind has introduced Genie 4, according to Google a new world model for real-time simulation. A world model is an AI that does not just output a finished video but keeps computing an environment frame by frame. You intervene, the world reacts.
The new part compared with its predecessor is sound. According to Google, every world generates ambient noise, music and dialogue on the fly. You steer in plain language, such as "camera to the left" or "make it a thunderstorm".
Everything it produces carries SynthID, Google's invisible watermark for AI content.
Who can use it
Currently: hardly anyone. Google is releasing Genie 4 first to a small group of selected trusted testers. Access is announced for Google AI Ultra subscribers in "Project Genie" and for early partners in the Gemini API and Google AI Studio. At the moment Google gives no dates or prices.
What does it cost? Unclear. No free offer is in sight, and Ultra is a paid subscription. If something cheaper arrives later, expect a usage limit rather than a free playground.
New? Not quite
Google calls Genie 4 the first real-time world model with native audio. Reports say a competing product from Runway already generates audio and is playable too. Whoever says "first" should be able to prove it. Here, the picture is disputed.
Reports on session length also differ: some say two minutes, others "a few minutes". That is not long. A world that frays after a short while is not ready for a whole game.
What it is for
The value lies less in playing than in trying things out. Game developers could sketch ideas in minutes. Robotics researchers could train AI systems in simulated environments before they break something in the real world. Everyone else watches in awe, just like you right now. With new AI techniques this is common: first the demo, then the waiting list, then eventually a product you can use without a waiting list. Until then, the technology behind it is worth a look.
Is this for you? For building: only if you get into the partner programs. For understanding where AI video is heading: yes, absolutely.
What you can take away today
You do not need a genie to grasp the principle. Ask your chatbot: "Describe a scene so that I could explore it in a game: place, sounds, three things I can interact with." You instantly see how much world fits into one sentence. Genie 4 turns it into one that reacts.
Access for more people will show how this goes beyond "impressive, but an early test". Until then: amazement allowed, miracles not expected.