AI RESEARCH
CatalogStitch: Dimension-Aware and Occlusion-Preserving Object Compositing for Catalog Image Generation
arXiv CS.CV
•
ArXi:2604.08836v1 Announce Type: new Generative object compositing methods have shown remarkable ability to seamlessly insert objects into scenes. However, when applied to real-world catalog image generation, these methods require tedious manual intervention: users must carefully adjust masks when product dimensions differ, and painstakingly re occluded elements post-generation. We present CatalogStitch, a set of model-agnostic techniques that automate these corrections, enabling user-friendly content creation.