Michel Oscar ; Bar-On Roi ; Liu Richard ; Benaim Sagie ; Hanocka Rana

Abstract

In this work, we develop intuitive controls for editing the style of 3Dobjects. Our framework, Text2Mesh, stylizes a 3D mesh by predicting color andlocal geometric details which conform to a target text prompt. We consider adisentangled representation of a 3D object using a fixed mesh input (content)coupled with a learned neural network, which we term neural style fieldnetwork. In order to modify style, we obtain a similarity score between a textprompt (describing style) and a stylized mesh by harnessing therepresentational power of CLIP. Text2Mesh requires neither a pre-trainedgenerative model nor a specialized 3D mesh dataset. It can handle low-qualitymeshes (non-manifold, boundaries, etc.) with arbitrary genus, and does notrequire UV parameterization. We demonstrate the ability of our technique tosynthesize a myriad of styles over a wide variety of 3D meshes.