Text2Texture: Generating 3D-Printed Models with Textures based on Text and Image Prompts

Abstract

To support users’ understanding of physical properties in 2D images, we propose Text2Texture, a webtool that converts 2D color images into textured 3D objects ready for 3D printing. This is achieved by extracting depth information using a monocular estimator, extracting local texture information using a fine-tuned stable diffusion model, and superimposing these macro- and micro-scale geometries to produce a composite 3D model with color, depth and texture. Images can be uploaded directly or generated via text prompt, and we print a variety of objects generated using each approach to suggest applications in physicallizing virtual worlds, adding haptic cues to photographs, and conveying information about scale in images.