From Sketch to 3D

This chapter outlines emerging workflows for generating three-dimensional architectural forms directly from sketches and concept images using ComfyUI. It introduces the principles of image-to-voxel inference, describing how models such as Hunyuan3D translate 2D sketches into latent volumetric representations before reconstructing meshes via voxel decoding. A step-by-step workflow is presented, including sketch preparation, background removal with REMbg, feature encoding through CLIP vision, volumetric conditioning, and final mesh export to .glb or .obj formats. Applications include massing studies, VR-ready concept models, and rapid site integration, enabling architects to bypass conventional CAD-to-3D sequences. The chapter also identifies technical constraints: non-metric accuracy, limited mesh resolution, and challenges with internal voids or complex geometries. Optional refinements such as remeshing, Boolean operations, and AI-driven retexturing are suggested as methods for extending base outputs into usable design assets. By repositioning sketches as computational seeds for volumetric generation, the chapter demonstrates how ComfyUI can compress traditional workflows and provide new opportunities for iterative, expressive, and collaborative design exploration in three dimensions.