jc1518 / image-reader

A project to explore various foundation models that have vision capabilities in Amazon Bedrock.

Image Reader

Introduction

A project to explore various foundation models that have vision capabilities in Amazon Bedrock.

• Image Reader uses Claude 3 multimodal models to interpret images or transcribe text in the images.
• Image Finder uses Titan multimodal embedding model to find the similar images by text or image.
• Image Library uses ChromaDB as the vector database for storing images embeddings.
• Image Generator uses Titan image generator model to generate images.

Requirements

• Use Python 3.11+, and install dependencies: pip install -r requirements.txt.

• Default bedrock region is us-west-2, change the value of BEDROCK_REGION in constant.py accordingly if you use other region.

• Request access to Claude 3 models and Titan models in Bedrock if you have not done that.

Use locally

Setup AWS credentials, then run cd image-reader; streamlit run Home.py

Deploy to AWS

Setup AWS credentials, then run

• Customize the config.yaml
• Install dependencies cd cdk; npm install
• Deploy npx cdk deploy --require-approval never

Demo

• Image-Reader
• Image-Finder

Blog

• Image-Reader: A project to explore Claude 3 Vision Capabilities
• Prompt Engineering with Claude 3 Haiku
• Unleashing the Power of Generative AI: Building a Smart Image Library