Google Genie: Create Interactive Worlds from Photos & Prompts
▼ Summary
– Google has released Project Genie, a more widely available version of its Genie 3 AI world model, but access is restricted to its most expensive AI subscription tier.
– These AI world models generate dynamic, interactive video environments that respond to user control inputs, simulating exploration rather than creating true 3D worlds.
– A key breakthrough of Genie 3 was its enhanced long-term memory, allowing it to maintain world details for a couple of minutes during generation.
– Users can create new worlds by providing a reference image or a text prompt describing the desired environment and character, a process Google calls “world sketching.”
– The system generates a 720p video at about 24 frames per second, rendering the path ahead in near real-time as the user moves a character.
Google’s Project Genie is now available to a broader audience, offering the ability to generate interactive, explorable worlds from simple text prompts or reference images. This tool represents a significant step in AI-driven world modeling, allowing users to sketch a concept and then navigate through a dynamic video simulation. Previously limited to a select group of testers, access is now tied to Google’s premium AI subscription tier, integrating with advanced models like Nano Banana Pro and Gemini 3.
The core technology is a refined version of last year’s Genie 3 model. While not creating true 3D environments, these world models generate responsive video sequences. You control a character, and the AI renders the surrounding scenery in real-time based on your movements, creating an immersive exploratory experience. A key advancement was the model’s enhanced long-term memory, enabling it to maintain consistency within the generated world for several minutes, a notable duration in this field.
Users can initiate the process through what Google terms “world sketching.” You start by providing a text description or an image as a creative foundation. The system, often utilizing Nano Banana Pro, produces a static scene. If this initial image doesn’t match your vision, you can edit or adjust it before handing it over to Genie. From that single frame, the AI builds the entire interactive environment.
The final output is a 720p video that runs at approximately 24 frames per second. Using standard controls like the WASD keys, you guide your character through the world. Genie renders the path ahead in near real-time, dynamically creating the landscape, objects, and details as you explore. This allows for experimentation with seemingly infinite variations, all generated from a simple starting point.
(Source: Ars Technica)
