TailorCode: 3D Clothed Human Generation from Images with Text-Instructed Editable Clothing Dimensions

3D virtual garment can be widely applicable in fields like digital human applications, fashion design, gaming industry, etc., where these models are typically built by highly experienced fashion modelers with extensive labor over a long period. Image-based and text-guided 3D model generation using deep learning networks achieves impressive results of cutting down on manpower and time. This paper presents an innovative approach for generating editable 3D clothed human models from a single image. Our method enhances existing garment reconstruction techniques and introduces a specialized measurement-editable module thereby establishing a direct connection between clothed human images and corresponding 3D garment models, enabling precise adjustments of clothing attributes such as length and style. Also, fine details like wrinkles are maintained through a finetuning structure. Our proposed solution demonstrates superior semantic guidance and detail restoration, effectively reconstructing both fitted and loose garments and editing 3D garment from monocular image, and also paves the way for downstream applications in animation, texturing, and simulation of clothed digital humans.