Google DeepMind has taken a significant step forward in AI-driven creativity.
With the launch of Project Genie, an experimental prototype that lets users generate and explore interactive virtual worlds on demand, Google introduces an AI model that goes way beyond mere text-to-video, effectively transforming the paradigm from video generation into real‑time, interactive world creation.
Powered by Genie 3, the tool has now moved from limited trusted tester access to a broader (though still restricted) rollout, marking a shift from static AI generations to dynamic, real-time simulations.
At its core, Project Genie transforms simple inputs, like text prompts or images, into fully navigable 3D environments.
Users can sketch out whimsical scenes, such as a castle built from marshmallows floating in the clouds with rivers of chocolate sauce, or more grounded ones based on uploaded photos of everyday spaces.
The system integrates with tools like Nano Banana Pro for image generation and refinement, and Gemini for overall orchestration, allowing quick previews, adjustments, and perspective choices (first-person or third-person views).
Once a world loads, exploration happens in real time: as you move using keyboard controls (WASD for navigation, spacebar for jumping or ascending), the environment generates ahead of you, maintaining consistency in objects, lighting, and physics where possible.
Genie 3 represents a major advancement over earlier versions.
Previous models could produce short, coherent video-like simulations, but Genie 3 introduces true interactivity and longer-term memory through auto-regressive generation.
It predicts how the world evolves based on actions, simulating realistic physics: objects fall, characters interact, and revisited areas stay largely stable.
Outputs run at 720p resolution and 24 frames per second, supporting diverse styles from claymation and watercolor to anime or near-photorealistic scenes.
While Genie introduces a whole new world of experiences, and that Google even touted it as "a key stepping stone on the path to AGI," results can be inconsistent.
Controls sometimes feel unresponsive, and worlds may glitch (like walking through walls). But still, the coherence over short sessions (up to about 60 seconds of generation per interaction) is impressive for an experimental tool.
The prototype emphasizes fun and experimentation rather than polished production.
Users can remix existing worlds, browse a curated gallery or randomizer for inspiration, and even download videos of their explorations.
Hands-on testers have crafted everything from candy-filled fantasy realms to mini-game-like experiences reminiscent of classic platformers, though safety guardrails block copyrighted or inappropriate content. Navigation can feel chaotic for non-gamers, and photorealistic or highly detailed prompts often yield more stylized, game-like results than lifelike ones.
Google also puts some strict filters, as well as guarails, which prevent Genie to generate nudity, or anything similar to some copyrighted material, most notably, Disney.
Back in December, Disney hit Google with a cease-and-desist, accusing Google's AI models of copyright infringement by training on Disney’s characters and IP and generating unauthorized content, among other thinggs.
According to Google in a blog post, access to Project Genie is currently limited to Google AI Ultra subscribers in the U.S. who are 18 or older, at a subscription cost of $249.99.
Interested users can head to labs.google/projectgenie to get started after subscribing. Google plans to expand availability to more regions based on feedback, with future updates likely addressing limitations like session length, latency, and adding features such as promptable events.
Positioned as a stepping stone toward artificial general intelligence, Project Genie highlights how world models could one day train AI agents in safe virtual spaces for robotics, education, game development, or autonomous systems testing.
For now, it offers a playful glimpse into an AI future where anyone can summon infinite, evolving worlds from a single idea, blurring the line between imagination and interactive reality.