Skip to main content

Image Embedder

Generate embeddings from images using AI models for similarity and clustering tasks.

What is the Image Embedder?​

The Image Embedder node converts images into numerical vector representations (embeddings) using AI models. These embeddings capture the visual features and semantic meaning of images, enabling tasks like image similarity search, clustering, and visual analysis.

How to use it?​

Configure the Image Embedder with these parameters:

1. Model Selection​

Choose from available image embedding models:

AWS Bedrock Models:

  • Amazon Titan Image Embeddings V1 (amazon.titan-embed-image-v1)
    • High-quality image embeddings for similarity and search tasks
    • Available regions: US East 1, US West 2, AP South 1, AP Southeast 2, CA Central 1, EU Central 1, EU West 1, EU West 2, EU West 3, SA East 1

Google Models:

  • Multimodal Embedding (multimodalembedding@001)
    • Supports both images and text in a unified embedding space
    • Available in multiple GCP regions worldwide

2. Credentials​

Provide the appropriate credentials based on your selected model:

  • AWS Bedrock Models: Requires AWS credentials or NoCodo Managed Credentials
  • Google Models: Requires Google Cloud credentials

3. Region​

Select the region where the model should be deployed:

  • AWS Bedrock: Choose from supported AWS regions (varies by model)
  • Google Cloud: Choose from supported GCP regions (global availability)

Output​

The Image Embedder outputs an Image Embedder instance that can be connected to Vector Store Readers and Writers.

Common Use Cases​

Create a search system that finds visually similar images:

  1. Generate embeddings for all images in your collection
  2. Store embeddings in a vector database
  3. Query with a new image to find similar matches

Best Practices​

  • Model Selection: Choose AWS Titan for pure image embeddings, or Google Multimodal for cross-modal (image + text) tasks
  • Region Selection: Select regions close to your data storage for lower latency
  • Batch Processing: Process multiple images in batches when possible for better efficiency
  • Vector Storage: Always store embeddings in a vector database for efficient similarity searches

Notes​

  • Image embeddings are numerical vectors that represent visual features
  • Different models produce embeddings of different dimensions
  • Embeddings from the same model can be compared using cosine similarity or other distance metrics
  • For best results, use the same model consistently across your image collection