n8n Image Embeddings Workflow: Text Summarisation Search

Find the right image instantly with n8n image embeddings—powered by Vision + vector search

This n8n template ingests photos from Google Drive, builds semantic image embeddings using both color metadata and vision-based descriptions, and then lets you query your image library like text. In short: type a phrase, and the workflow returns the relevant image references and/or links from your vector store.

What this workflow does

• Imports a photo from Google Drive into the workflow.
• Uses the Edit Image node to extract colour information, which becomes part of the image’s semantic metadata.
• Processes the same photo with a vision-capable model to generate a short description and semantic keywords.
• Merges the colour metadata + vision output with the document metadata to create a single “image document”.
• Inserts that document into a vector store as a text embedding associated with the image.
• Enables querying the vector store like a normal document search—returning the relevant image references/links.

Use cases

• Personal photo library search: find images by meaning (e.g., “sunset beach”, “red car”).
• Product recommendations: retrieve visually similar products using semantic queries.
• Video footage exploration: index frames or extracted images and search via descriptive prompts.

Technical details

• Integrations / nodes: set, merge, edit image, sticky note, Google Drive, manual trigger.
• AI requirements: an OpenAI account for the vision-capable model and embedding models.
• Storage: image documents are stored as text embeddings in a vector store to support contextual retrieval.

Note: The workflow is designed for creating image embeddings via text summarisation + metadata. If you want to improve results, you can swap in dedicated multimodal embedding approaches (e.g., Google Vertex AI multimodal embeddings).