EDITING FASHION IMAGES WITH PRECISION: A CONTROLLED IN PAINTING METHOD

Text-guided image editing, especially in the fashion domain, is a powerful yet complex task that involves modifying specific elements of an image based solely on textual instruction. The system must not only interpret the instruction but also localize and edit the relevant image regions while preserving the rest of the visual content. In this article, we present a practical and controlled methodology for generating a paired dataset to support this task. Our approach builds on recent advancements in generative AI, combining Stable Diffusion inpainting techniques with pre-trained large language models to produce consistent and faithful edits. Specifically, we leverage human parsing and fashion-specific datasets to enable localized garment transformations based on color and fabric changes. This article complements the work presented in our paper” AI-Powered Text-Guided Image Editing: Innovations in Fashion and Beyond”, presented at the ISPR 2025 conference, by providing a focused exploration of the dataset generation pipeline and implementation details used in that study