How AI-Powered Virtual Try-On Transforms Fashion E‑Commerce

1 Introduction

Traditional online shopping often leaves consumers unable to intuitively gauge how clothing will look on them, leading to high return rates. JD.com’s AI try‑on solution Oxygen Tryon lets users upload a photo and instantly see how garments fit, displaying shape, color, and texture, thereby enhancing satisfaction and lowering returns.

2 Background

2.1 Business Scenario

In fashion e‑commerce, shoppers rely on static images and text, which often misrepresent the actual fit, resulting in return rates of 30‑40%. High returns increase logistics costs, product loss, and customer‑service pressure, while limiting conversion growth.

2.2 Technical Challenges

Since 2015, the industry has explored AI try‑on, initially using GAN‑based image synthesis, which could not simulate body motion, fabric drape, or realistic wrinkles. Later approaches combined real‑time body tracking with AR 2.5D overlays, improving interaction but still lacking realism, complex pose handling, and lighting consistency.

3 Technical Implementation

Oxygen Tryon first extracts human key points from a model, creates masks for garment regions, and feeds these into a Redux model for feature extraction. The features become prompt embeddings for the Fluxfill model, which generates the final try‑on image.

3.1 Technical Innovations

Accurate Body Recognition: JD’s proprietary algorithm precisely measures body dimensions, ensuring garments drape naturally and fit each body shape.

Realistic Material Rendering: High‑fidelity rendering of fabrics such as cotton, silk, and denim captures reflections, refractions, wrinkles, and texture for a lifelike appearance.

Fast Generation: The system produces high‑definition try‑on images within 7 seconds, providing a smooth, low‑latency experience.

Intelligent Outfit Recommendation: A multimodal model suggests complete outfits and supports full‑body try‑on based on user’s skin tone, hairstyle, and facial features.

3.2 Generated Results

4 Future Outlook

Oxygen Tryon has significant growth potential. JD plans to partner with more fashion brands, aiming for over 30 brands and 100,000+ SKUs to support AI try‑on during the 11.11 shopping festival.

Algorithmic upgrades will focus on maintaining consistency for complex patterns, logos, and textures, and on realistic simulation of fabric physics under gravity and friction.

Future features include personalized recommendations based on skin tone, hairstyle, and facial features, integration with size‑assistant tools for a complete decision chain, and expansion to other categories such as shoes, jewelry, glasses, and accessories.

With continued team effort, Oxygen Tryon is expected to keep leading fashion e‑commerce innovation.