According to engadget, Google DeepMind is now making its Genie 3 AI model available to people outside of Google through Project Genie. To access it, you’ll need to subscribe to Google’s $250 per month AI Ultra plan, be located in the US, and be 18 or older. The launch offers three modes: World Sketching, exploration, and remixing, using the Nano Banana Pro model to generate a source image. Users can describe a character, define a camera perspective, and preview a “sketch” before Genie 3 builds the interactive world. However, these AI-generated worlds are capped at 60 seconds, 24 frames per second, and 720p resolution. DeepMind initially positioned Genie 3 as a tool for training AI agents when it debuted this past summer.
What is Genie, really?
Here’s the thing: it’s crucial to understand what this is not. Genie 3 is not a game engine. It doesn’t have points, objectives, or complex mechanics. It’s more like a physics-aware, interactive diorama generator. You give it a starting point—either a text prompt or an image—and it simulates a consistent, navigable space for one minute. The fact that it can maintain coherence and react to user input (like moving left or jumping) is the real technical magic. It’s simulating a world model on the fly, which is a foundational step toward more general AI. But for now, you’re basically getting a tech demo of a very smart, very expensive screensaver.
The access problem
Now, let’s talk about that $250 per month price tag. That’s a massive barrier to entry for curious tinkerers and indie creators. It immediately positions Project Genie as a tool for professionals, researchers, or the seriously dedicated. Google is clearly using the AI Ultra tier as a filter, limiting the compute load to a small, paying audience while they test. It makes business sense, but it also means the most interesting and weird creations from a broad community won’t happen yet. And with the 60-second, 720p limit, you have to wonder: is the current output valuable enough to justify that cost for anyone outside of a lab? Probably not for most folks.
Why this matters
So why is DeepMind doing this? The big picture is about training AI. A model that can understand and generate interactive worlds is a model that understands cause, effect, and physics. That’s incredibly valuable for creating more capable AI agents that can operate in simulated environments before touching the real world. By opening it up, even in a limited way, they’re crowdsourcing stress tests and finding creative use cases they hadn’t considered. It’s a research project dressed in a slightly more consumer-friendly wrapper. The official blog post frames it as a creative tool, but the core ambition is likely far more technical. This is a peek at the infrastructure being built for the next generation of AI, and you’re paying for the privilege of being a beta tester on the absolute bleeding edge.
