Worldbuilders
Gemini 4 NYC HackathonWorld IR is an application that converts text prompts, images, and videos into interactive, editable 3D environments using Gemini AI models. The project addresses the limitations of traditional media, which is generally fixed to a particular viewpoint and difficult to modify or explore. Creating a 3D environment from existing images or videos typically requires manual modeling, animation, and scene reconstruction. World IR uses Gemini 4 to interpret input content and generate a structured description of the environment, including objects, spatial relationships, materials, lighting, and movement. The application converts this information into a persistent 3D scene that users can explore using free camera controls. For video inputs, the system uses computer vision to track moving objects and reconstruct their approximate trajectories in 3D. For image inputs, it uses visual references to guide scene layout and appearance. Text prompts can generate environments without requiring reference media. Users can navigate the resulting environment, change camera viewpoints, follow moving objects, and modify individual scene elements using natural-language instructions. Changes are applied to the existing scene without rebuilding unrelated objects. The project demonstrates how multimodal AI can convert static and recorded media into reusable, interactive 3D content for creative prototyping, visualization, education, and virtual production.





