Stylizing 3D Models With Generative AI for Fabrication

Abstract

This thesis presents two novel approaches for modifying 3D models using generative AI for stylization while ensuring the resulting models preserve the properties required for fabrication. The first method, Style2Fab, separates functional and stylistic sections of 3D models to enable targeted modifications that preserve the model's intended functionality. By distinguishing between these sections, Style2Fab allows for alterations that maintain the model's functional purpose while providing flexibility in its aesthetic design. This approach ensures that the modified models retain their original functionality after stylistic changes.

The second method, MechStyle, incorporates finite element analysis (FEA) into the generative modeling pipeline to maintain the structural integrity of the modified models. By analyzing changes in stress values during a simulated drop test at various stages of the stylization process, MechStyle restricts changes to those that preserve the model's structural viability. This ensures that the resulting models are both stylistically accurate to the user's desired results and structurally sound for 3D printing.