How AI-Powered Virtual Try‑On Is Revolutionizing Fashion E‑Commerce

1 Introduction

Traditional online shopping suffers high return rates because consumers cannot see how clothes look on them. JD.com’s AI try‑on "Oxygen Tryon" lets users upload a photo and instantly see a realistic preview of garment fit, texture, and color, improving satisfaction and reducing returns.

2 Background

2.1 Business Scenario

In e‑commerce, static images limit consumers’ ability to judge fit, leading to 30‑40% return rates and high operational costs for merchants.

2.2 Technical Challenges

Since 2015, AI try‑on evolved from GAN‑based static synthesis to real‑time 2.5D models, yet challenges remain in body motion, realistic fabric rendering, complex scenes, and lighting.

3 Technical Practice

Oxygen Tryon extracts human key points, masks garment regions, uses a Redux model for feature extraction, and feeds prompts into the Fluxfill model.

3.1 Innovations

Accurate body measurement : JD’s algorithm analyzes body dimensions to ensure natural drape and fit.

Realistic material rendering : High‑precision rendering of fabrics (cotton, silk, denim) with correct light interaction and wrinkles.

Fast generation : High‑definition try‑on images are produced within 7 seconds.

Smart outfit recommendation : Multimodal model suggests complete outfits and supports full‑body try‑on.

3.2 Generated Results

4 Future Outlook

Future plans include expanding brand partnerships (over 30 brands, 100k+ SKUs for JD’s 11.11 event), improving consistency for complex patterns, enhancing physical simulation of fabrics, adding personalized styling based on skin tone and facial features, and extending try‑on to shoes, jewelry, glasses, and accessories.

With continued effort, Oxygen Tryon aims to lead fashion‑e‑commerce AI innovation.