AI-Powered Product Catalog Revolution: From Photos to Rich Listings

This sample project demonstrates an AI-powered solution that transforms simple product photos into rich, detailed product listings using Amazon Bedrock's foundation models. The system automatically generates product descriptions, creates virtual try-on images, and provides comprehensive product attribution.

Quick Demos

Product Catalog Demo

Attribution Demo

Architecture Overview

The architecture diagram illustrates the following processing workflow:

1. User Interaction: Users access the Streamlit web application running on EC2 to upload product images and configure generation parameters (gender, pose, body structure, emotion, brand voice, etc.)

2. Image Upload & Storage: Product images and optional human model images are uploaded and stored in Amazon S3 buckets for processing

3. Workflow Orchestration: AWS Step Functions initiates and orchestrates the AI processing workflow, managing the sequence of operations and error handling

4. Product Attribution: The first Lambda function (product-attribution) analyzes the uploaded product image using Amazon Bedrock's Nova Pro model to:

   • Extract product features and attributes
   • Generate detailed product descriptions
   • Apply brand voice and pricing influences based on user parameters

5. Virtual Try-On Generation: The second Lambda function (image-try-on) leverages Amazon Bedrock's Nova Canvas model to:

   • Generate or use the provided human model image
   • Apply the product to the human model based on specified parameters (pose, body structure, emotion)
   • Create realistic virtual try-on images

6. Data Persistence: Product information, metadata, and processing status are stored in Amazon DynamoDB for tracking and retrieval

7. Output Storage: Generated images (virtual try-on results) and product descriptions are stored back in Amazon S3

8. Results Display: Users can view the generated content through the Streamlit interface, accessing both the AI-generated product descriptions and virtual try-on images

Features

• Product Image Upload: Upload product photos for AI analysis
• Virtual Try-On: Generate realistic images of products on human models
• Customizable Parameters: Configure gender, pose, body structure, emotion, brand voice, and pricing influences
• AI-Generated Descriptions: Automatic product attribution and description generation from any set of photos. Example use cases below:
  • Generate hotel room listing from pictures of Room.
• Secure Deployment: VPC-isolated EC2 instance with SSM access

Prerequisites

Before deploying this solution, ensure you have:

1. AWS CLI configured with appropriate permissions
2. AWS CDK installed (npm install -g aws-cdk)
3. Node.js (version 18 or later)
4. Docker installed and running
5. jq command-line JSON processor (Optional only for local development)
6. AWS Account to deploy this sample project

Enable Amazon Nova models

Before deployment, you must enable the required foundation models in Amazon Bedrock:

1. Navigate to the Amazon Bedrock console
2. Go to "Model access" in the left navigation
3. Request access to:
   • Amazon Nova Pro (amazon.nova-pro-v1:0) - for text generation
   • Amazon Nova Canvas (amazon.nova-canvas-v1:0) - for image generation

Refer Amazon Nova Canvas update: Virtual try-on and style options now available

Deployment Instructions

1. Clone and Setup

git clone https://github.com/aws-samples/sample-ai-powered-product-catalog.git
cd sample-ai-powered-product-catalog
npm install

2. Bootstrap CDK (if not already done)

cdk bootstrap

3. Deploy the Stack

cdk deploy AutomatedProductCatalogStack

The deployment will create:

• All AWS infrastructure components
• A containerized Streamlit application running on EC2
• VPC with appropriate security groups and endpoints

Accessing the Application

Using UI from EC2 via System manager Port forwarding

The EC2 instance is deployed in a private isolated subnet without internet access. Use AWS Systems Manager Session Manager to create a secure tunnel:

1. Start SSM Port Forwarding Session:

   # Get the EC2 instance ID from the AWS Console or CLI
   INSTANCE_ID=$(aws ec2 describe-instances \
     --filters "Name=tag:aws:cloudformation:stack-name,Values=AutomatedProductCatalogStack" \
               "Name=instance-state-name,Values=running" \
     --query 'Reservations[0].Instances[0].InstanceId' \
     --output text)
   
   # Create port forwarding session
   aws ssm start-session \
     --target $INSTANCE_ID \
     --document-name AWS-StartPortForwardingSession \
     --parameters '{"portNumber":["8501"],"localPortNumber":["8501"]}'

2. Access the Application: Open your browser and navigate to: http://localhost:8501

3. Stop the Session: Press Ctrl+C in the terminal where the SSM session is running

Developing locally or running UI locally

For development and testing, you can run the Streamlit application locally:

1. Navigate to the UI directory:

   cd ui

2. Install Python dependencies:

   pip install -r requirements.txt

3. Run the local development script:

   chmod +x start-local.sh
   ./start-local.sh

   This script will:

   • Set the AWS region to us-east-1
   • Fetch CloudFormation stack outputs and set them as environment variables
   • Start the Streamlit application locally

4. Access the Application: The script will display a local URL (typically http://localhost:8501)

Using the Application

1. Upload Product Image: Select a product photo (JPG, JPEG, or PNG)

2. Optional Human Model: Upload a human model image, or let AI generate one

3. Configure Parameters:

   • Gender: male, female, or unisex
   • Pose: Straight, Hand raised, or Looking aside
   • Body Structure: Average (Build A), Oversize (Build B), or Thin (Build C)
   • Emotion: Confident, Amazed, or Funny
   • Brand Strength: Fabric, Style, Design, or Competitive Pricing
   • Brand Voice: Budget, Fun, or Trustworthy
   • Pricing Influence: Inventory Level, Fabric, or Weather
   • Promoted Product: Toggle for promotional treatment
   • Number of Output Images: 1-5 generated images

4. Submit: Click Submit to start the AI processing workflow

5. View Results: Navigate to the outputs page to see generated images and product descriptions

Cleanup

To avoid ongoing AWS charges, destroy the stack when no longer needed:

cdk destroy AutomatedProductCatalogStack

Costs

Running this sample for processing 1000 images per month costs about 170$ to 290$ in us-east-1 region.

• Amazon Bedrock - Amazon Nova Canvas for AI generated human images 3 x 1000 = 3000 x 0.04 = 120$ (This applies only when not using actual huamn model images)
• Amazon Bedrock - Amazon Nova Canvas for Virtual tryon images 3 x 1000 = 3000 x 0.04 = 120$
• Amazon Bedrock - Amazon Nova Pro for Attribution
• Amazon EC2 36$
• Amazon DynamoDB 3$
• AWS Step function 0.5$
• Amazon S3 1$ (15g storage)
• AWS Lambda 1$

License

This library is licensed under the MIT-0 License. See the LICENSE file.

Contributing

See CONTRIBUTING for more information on how to contribute to this project.

Security

See CONTRIBUTING for more information about reporting security issues.