Google launches Project Genie AI prototype

Google has launched Project Genie, an experimental web-based prototype powered by Genie 3 (with Nano Banana and Gemini) that uses World Sketching to turn text or images into dynamic, explorable 3D worlds which users can create, remix and navigate in real time. Access is currently limited to Google AI Ultra subscribers in the U.S., sessions are capped at 60 seconds, and Google warns the tool has notable limitations in realism, physics and control even as it highlights potential uses in education, animation and simulation-based training.

Google has launched Project Genie, an experimental web-based prototype powered by Genie 3 (with Nano Banana Pro and Gemini) that generates interactive 3D worlds from text prompts or images and is rolling out to Google AI Ultra subscribers in the United States. Users create scenes via a World Sketching feature (or by uploading images), then enter World Exploration to move through environments in first- or third-person and can remix others’ worlds with new prompts. Google and DeepMind present the project as part of research into world models and AGI, emphasizing that the system dynamically generates the path ahead rather than producing static renders, with potential applications in education, training, animation, and design. The release is explicitly experimental: sessions are limited to 60 seconds, generated worlds can look rough, physics and controls may be inconsistent, and initial access is restricted to Google’s premium AI Ultra tier. Google frames the rollout as a shift from private research previews to broader public testing to learn how people interact with AI-created, navigable environments despite current shortcomings.

Key Points

Topics