Abstract: This paper presents DM-VTON, a diffusion-model-based virtual try-on (VTON) system designed to generate high-quality, photo-realistic visualizations of clothing on human subjects. The system simulates how garments fit and drape on the human body under real-world conditions, accounting for diverse poses, lighting variations, and body shapes. It integrates advanced artificial intelligence techniques including diffusion models, IP-Adapters, DensePose, and semantic segmentation to achieve accurate and realistic garment rendering. The model was trained on benchmark datasets VITON-HD and DressCode, enabling robust handling of both paired and unpaired try-on scenarios. Key contributions include dense pose mapping, agnostic image generation, and a Gradio-based user interface supporting high-resolution inference at 1024×768 pixels. The modular and scalable architecture effectively tackles persistent challenges such as occlusion handling, body shape variation, and lighting inconsistency. This work advances the fields of virtual try-on, fashion synthesis, and AI-driven apparel visualization, with significant applicability to e-commerce, fashion design, and sustainable retail.
Keywords: Virtual Try-On, Diffusion Models, DensePose, IP-Adapters, Semantic Segmentation, Fashion Synthesis, E-Commerce, Sustainable Fashion, Agnostic Image Generation, Garment Rendering.
Title: AI-Powered Virtual Outfit Try-On System for Sustainable Fashion and Sizing Accuracy
Author: Nishchay Sinha, Pinank Trivedi, Yashraj Verma, Deepak S. Shete
International Journal of Electrical and Electronics Research
ISSN 2348-6988 (online)
Vol. 14, Issue 2, April 2026 - June 2026
Page No: 1-9
Research Publish Journals
Website: www.researchpublish.com
Published Date: 09-April-2026
