How AI‑Driven Virtual Try‑On Boosted Fashion Sales by 80%

1. Introduction

In fashion e‑commerce, the main product image directly influences click‑through and conversion rates. Traditional photo shoots with real models are costly and time‑consuming, and rely on subjective experience, leading to a gap between product appeal and sales performance.

JD Retail Technology’s “Jing Dian Dian” AI try‑on combined with the Jingmai A/B testing platform uses AIGC to automatically generate realistic model images and short videos for apparel, achieving up to an 80% increase in order conversion during the 618 promotion.

2. Background

2.1 Business Scenario

Fashion apparel requires complex material creation involving model booking, scene setup, shooting, and post‑production, resulting in high costs and long cycles (up to a month for a new SKU). Small merchants often resort to flat or low‑quality images, reducing main‑image attractiveness.

Traditional studio shoots produce images that may not match the preferences of the actual buying audience, making it difficult for fashion merchants to leverage A/B testing for main‑image optimization.

The “Jing Dian Dian” platform (https://ai.jd.com/) offers AI‑driven try‑on that can batch‑generate diverse, realistic model images and 10‑15 second videos showing the garment from multiple angles, enabling rapid conversion‑rate optimization during large‑scale promotions.

2.2 Technical Challenges

The AI try‑on system faces several challenges:

Realism of generated models : Skin texture, hair detail, and lighting often appear overly smooth (“oil‑slick” effect), and body parts may be distorted or missing.

Clothing detail and style fidelity : Difficulties in accurately reproducing garment silhouettes, textures, colors, and small details such as buttons, zippers, lace patterns, and logos.

Adaptation to different models : Maintaining natural drape and fit across various poses, body types, and sizes.

3. Technical Practice

3.1 Innovations and Implementation

To address the challenges, “Jing Dian Dian” introduced several innovations:

Iterative virtual model library : A high‑quality library covering multiple ethnicities, genders, body types, and poses, enhanced with prompt‑engineering and a LoRA model for realistic skin.

Core algorithms : A 12‑billion‑parameter Flux‑Fill diffusion model (Transformer‑based) combined with dwpose pose estimation and parsing for sub‑pixel keypoint accuracy and fine‑grained segmentation.

Feature extraction : A Redux model extracts visual features from input garments, which are injected as priors into the inpainting process to preserve texture, pattern, and color.

Dynamic adaptive mask : Automatically adjusts mask generation based on garment category, ensuring stable and realistic try‑on across different styles.

These techniques enable one‑click “see‑what‑you‑get” experiences.

3.2 Results

In a case study with a male‑apparel KA brand during the 618 event, the AI system generated multiple style‑specific main images and videos (e.g., business elite, casual social, mature stable). Compared with a manually produced baseline, the AI‑generated “mature stable” style increased order conversion by over 80%.

Cost per SKU image/video : Traditional ¥2,000‑¥20,000, ≥15 days; AI ¥0, ≤0.5 day – cost reduced 100%, efficiency up 95%.

Main‑image A/B testing : Traditional ≥30 days, manual; AI ≤1 day, batch – efficiency up 30×.

Technical integration complexity : Traditional required multiple teams; AI can be operated by a junior staff online – barrier lowered >95%.

Key outcomes include:

First‑round elimination: Traditional images lagged by 66.67% in conversion within 72 hours.

Final victory: AI‑generated “mature stable” images matched product attributes and consumer expectations, boosting sales by 80%+.

Operational decision evolution: Data‑driven selection replaced experience‑based choices, improving main‑image selection efficiency by 30×.

Traffic agility: 24/7 automatic optimization maximized conversion during peak traffic.

4. Future Outlook

Personalized AI try‑on for thousands of user profiles, covering age, body shape, skin tone, and special groups.

Fully automated A/B testing that launches experiments and selects winning assets without manual intervention.

Predictive AI generation using conversion‑rate forecasting models to proactively create optimal main‑image combinations.