Enhancing Retail Experience: Implementing Virtual Try-On Technology
In this first post of a two-part series, we will delve into how retailers can adopt virtual try-on solutions to elevate customer experiences. Stay tuned for part two, where we will explore real-world applications and the benefits of this innovative technology.
Enhancing Customer Experience with Virtual Try-On Technology
In this first post of a two-part series, we delve into how retailers can effectively implement virtual try-on solutions to elevate customer experience. In our next installment, we’ll explore real-world applications and the myriad benefits of this cutting-edge technology.
The Returns Dilemma
As we move towards 2024, the fashion industry is grappling with a significant challenge: every fourth piece of clothing bought online is returned. This phenomenon contributes to an astonishing $890 billion returns problem in America. At the heart of these returns lies a simple yet crucial truth—shoppers struggle to gauge fit and style through their screens. Poor fit, incorrect size, or style mismatches rank among the leading reasons for returned fashion items.
Retailers face a unique dilemma, as their most valuable customers often return the highest number of items. This necessitates maintaining generous return policies, which can be costly and environmentally detrimental. Each return generates 30% more carbon emissions than the initial delivery and represents a missed sales opportunity until items are processed back into inventory. As online shopping accelerates, virtual try-on technology has emerged as a potential solution to reduce returns while enhancing customer convenience.
Overcoming Implementation Challenges
Early implementations of virtual try-on technology faced issues related to accuracy, scalability, and retaining essential garment details such as draping, patterns, and logos. To address these challenges, Amazon Nova Canvas has developed a robust virtual try-on capability that utilizes two-dimensional image inputs: a source image showcasing a person (or space) and a reference image of the product.
This system offers automatic product placement through auto-masking functionality as well as manual controls for precise adjustments. Crucially, it preserves important details like textures and logos, while providing comprehensive styling controls for customization.
Virtual Try-On in Action
Virtual try-on technology can be deployed across various customer engagement channels, such as ecommerce websites, mobile shopping apps, in-store kiosks, social media shopping platforms, and virtual showrooms. Imagine visiting an ecommerce site, uploading your personal image, and visualizing how clothing and accessories would look on you.
Exploring Amazon Nova Canvas
In this post, we’ll focus on the virtual try-on capability available in Amazon Nova Canvas. Retailers and ecommerce companies can easily integrate garment and product visualization into their customer touchpoints. By simply uploading a photo and selecting a product, customers can see how items would look on them—or even on a model—ushering in a new era of shopping convenience.
To experiment with virtual try-on in Amazon Nova Canvas, you can access it within the Amazon Bedrock playground. We’ll also provide guidance on implementing a complete solution around this feature in your Amazon Web Services (AWS) environment.
Solution Overview
At its core, the virtual try-on capability in Amazon Nova Canvas uses advanced AI processing to deliver real-time applications suitable for ecommerce. It maintains high-fidelity details of reference items, ensuring accurate semantic manipulations within scenes. Our solution employs AWS serverless services in an event-driven architecture, combining Amazon DynamoDB Streams, AWS Step Functions, and Amazon S3 to manage result delivery efficiently.
Detailed Architecture
-
Request Initialization
- Customer uploads their model photos and product images to Amazon S3.
- Each upload triggers an Amazon Simple Queue Service (SQS) message, prompting AWS Lambda functions to create metadata and store it in a DynamoDB product table.
- Amazon API Gateway manages WebSocket connections for real-time updates.
-
Virtual Try-On Generation
- A Step Function orchestrates the virtual try-on generation in coordination with DynamoDB to maintain job status and product information.
- The model uses auxiliary inputs for mask generation, ensuring precise placement of garments on the customer’s image.
-
Result Delivery
- Generated try-on images are stored in Amazon S3 with metadata linked to job IDs.
- SQS handles notifications for completed images, allowing AWS Lambda to send results back to users in real time through WebSocket connections.
Implementation Steps
When a customer submits a try-on request, they authenticate using Amazon Cognito and upload their photos to Amazon S3. A WebSocket connection through API Gateway is established for ongoing status updates. AWS Lambda processes the product request, manages entries, and retrieves product images for the try-on process.
The actual process involves generating a mask for the clothing (whether it be upper body, lower body, full body, or footwear) based on the specified garment type. Once the mask is created, the system overlays the product image onto the customer’s uploaded photo, generating the final try-on image—a process that typically spans just 7–11 seconds.
Conclusion
In this post, we’ve provided an overview of how to implement virtual try-on capabilities effectively. The growing demand for seamless and satisfying shopping experiences highlights the necessity of reducing return rates while maintaining consumer satisfaction. This virtual try-on solution exemplifies how AWS serverless services can be combined with generative AI to tackle a significant challenge in the retail sector.
Stay tuned for the second part of our series, where we will dig deeper into the real-world applications and extensive benefits of virtual try-on technology. In the ever-evolving landscape of ecommerce, being at the forefront of innovation can mean the difference between thriving and merely surviving.
About the Authors
Amandine Annoye is a Solutions Architect at AWS and works closely with Luxury & Fashion customers in France. Outside of her technical expertise, she enjoys traveling and painting.
Kevin Polossat is also a Solutions Architect at AWS and focuses on retail and consumer-packaged goods. He spends his free time indulging in wine and cheese.
Leopold Cheval works with Media & Entertainment and Retail customers as a Solutions Architect based in Paris. He enjoys travel and camping in his spare time.
Rania Khemiri serves as a Prototyping Architect at AWS, specializing in generative AI applications and empowering teams to transform their ideas into tangible prototypes.