Google DeepMind’s Project Genie: Create 3D Worlds with AI

• Google DeepMind has unveiled Project Genie, a tool capable of generating interactive 3D worlds from simple text prompts.
• The technology combines three AI models, enabling 60-second sessions within user-created environments.
• Starting in late January 2026, Google AI Ultra subscribers in the U.S. will be able to access the experimental platform.

The line between science fiction and reality is blurring as artificial intelligence advances. Google DeepMind’s Project Genie represents a significant leap forward, allowing users to create immersive 3D worlds simply by describing them. This innovation has the potential to reshape fields from design and training to gaming and entertainment, offering businesses new avenues for visualization and engagement.

Three Models, One Vision: How Project Genie Works

At the heart of Project Genie lies the interplay of three specialized AI systems. Genie 3, the World Model introduced in August 2025, forms the core of the technology. It simulates environments, calculates changes in state, and generates the world in real-time as the user explores. The system renders visuals at a resolution of 1280×720 pixels, achieving frame rates of 20 to 24 frames per second. Complementing Genie 3 is Nano Banana Pro, the image generator released in November 2025, which translates text inputs into visual designs and transforms sketches into photorealistic 3D objects. Finally, Gemini handles natural language processing, converting user descriptions into actionable instructions for the other two systems.

The workflow is designed to be intuitive. Users begin by describing their desired world – for example, a “Marshmallow Castle in the Clouds” or a “Claymation Wonderland with Chocolate Rivers.” They then select an avatar, choosing to embody a person, animal, or object. The perspective determines the experience – first-person, third-person, or movement via walking, flying, or driving. The system generates an initial draft, which users can refine before final generation.

Physics Meets Fantasy: What the Worlds Can Do

The generated environments are more than just visually appealing backdrops. Genie 3 simulates fundamental physics principles, including gravity, collisions, and material properties. Objects behave predictably – a ball rolls downhill, water flows, and structures respond to movement. While not every interaction is perfectly predictable, the consistency is notable. “Genie 3 generates the path ahead in real time as you move and interact with the world,” the Google DeepMind team explained. This real-time calculation distinguishes the system from passive video generators like OpenAI’s Sora.

Limitations of the Technology: Where It Still Needs Work

The 60-second session limit per use is a practical constraint. Each user requires a dedicated chip for the calculations, making it a resource-intensive process. “The reason we limit it to 60 seconds is because we wanted to bring it to more users,” Ori Fruchter of DeepMind told TechCrunch. Character control remains challenging, and not all physical interactions mirror reality. Latency issues can occur, particularly in complex scenes. Google emphasizes the experimental nature of the project, clarifying that it is a research prototype designed to evolve with user feedback, rather than a fully realized AGI application.

Access: A Premium Subscription as Your Ticket

Project Genie will launch exclusively for Google AI Ultra subscribers in the U.S. starting in late January 2026. The subscription costs $250 per month and includes access to Genie, higher usage limits, 30 terabytes of cloud storage, and the Antigravity Coding Tool. Interested users can join the waitlist through Google Labs. International expansion is planned, along with longer session times and potentially an API for developers. A full release is anticipated in 2026, with beta access prioritized for developers and creatives.

Practical Applications Beyond Gaming

The potential applications extend far beyond entertainment. In robotics, the technology can create training environments where machines learn complex movements without consuming real-world resources or incurring risks. Animation studios and film productions can quickly visualize scenes and explore variations before committing to expensive production resources.

Historical reconstructions become tangible: imagine walking through ancient cities reconstructed based on archaeological data. In education, abstract concepts can be translated into interactive experiences. For product developers, the technology offers the ability to test prototypes in various environments and provide stakeholders with immersive demonstrations long before a physical prototype exists.

Position in the Competitive Landscape: Interaction Over Passivity

While OpenAI’s Sora generates impressive videos, users remain passive observers. Project Genie transforms users into active participants within self-created worlds. The combination of real-time navigation and physics simulation sets the system apart from pure video generators. Competition exists from companies like World Labs and Runway, which are also developing interactive AI worlds. Google is positioning itself as a major player in this emerging market by integrating the technology into its existing ecosystem and leveraging the computational power of DeepMind.

The Path to Artificial General Intelligence

World Models like Genie 3 are considered crucial building blocks on the road to AGI. By learning how worlds function – how objects interact, how cause and effect are linked – they develop a fundamental understanding of physical reality. This capability is essential for systems designed to operate in the real world. In this context, Project Genie is more than a creative tool; it’s a testing ground for technologies that could one day control robots, conduct complex simulations, or test scientific hypotheses.

Remix and Gallery: Collaborative Creation

The platform allows users to adapt and build upon existing worlds. A curated gallery showcases creations from others, providing inspiration and a starting point for individual variations. A randomizer generates unexpected combinations, potentially overcoming creative blocks. At the end of each session, users can download a video of their exploration, making the results shareable and documenting the 60-second experience.

What This Means for Your Business

The technology is still in its early stages, but the potential is clear. Consider where spatial visualization, simulation, or interactive presentation could make a difference within your organization. The $250 monthly cost is manageable for businesses if it leads to efficiency gains or new customer experiences. Early adopters can gain valuable experience before the market matures and competitors emerge. The planned API will enable integration into existing workflows. Businesses that begin exploring the possibilities now will gain a competitive advantage.