Enhancing E-Commerce with Virtual Try-On Using Stable VITON
R. Raja Subramaniam, Machireddy Dhamini, Marni Srija, M Vaishnavi, Malepati Vidyadhari · 2025
We have seen a huge growth of popularity for online clothes shopping, the absence of a personalized try-on experience continues to affect customer satisfaction and purchase confidence. To address this challenge, we propose a deep learning-based virtual try-on system that enables customers to visualize how a selected outfit would appear on them before making a purchase. Customer can exchange their own image along with a screenshot of a desired clothes collected from popular e-commerce platforms such as Myntra or Flipkart. The system utilizes Stable VITON, a diffusion-based virtual try-on framework built upon Stable Diffusion v1.4, which generates high-quality, realistic try-on images by learning fine-grained garment alignment and human-body context. This model combines a U-Net-based denoising network, variational autoencoder (VAE), and a Zero Cross-Attention Block to precisely synthesize the outfit onto the person’s image. To enhance the output, attention enhancing techniques, defect removal modules, and attention total variation loss are applied, resulting in realistic try-on results with minimal artifacts. Unlike conventional recommendation systems, our model provides a direct visual interface for customer interaction, link the gap between online browsing and in-store trial experiences. The proposed system aims to redefine fashion e-commerce by making virtual fitting more realistic, accessible, and customer-centric.