Generating Image Embeddings via Textual Summarisation

This n8n workflow enables quick image contextual search by converting images into searchable text embeddings. It processes photos from Google Drive, extracts visual metadata like colors, and uses a vision AI model to generate semantic descriptions, combining them into a document embedded in a vector store.

The workflow starts with a manual trigger or image import from Google Drive. An Edit Image node extracts color information for semantic metadata. A vision-capable OpenAI model analyzes the image to produce a short description with keywords. These elements merge with image metadata to create a comprehensive text document, which is then embedded and stored in a vector database, linking back to the original image.

Benefits include efficient image search without dedicated multimodal models, saving costs and setup time. Use cases span personal photo libraries for quick retrieval, e-commerce product recommendations based on visual similarity, and video footage analysis by sampling frames. It democratizes advanced search for non-technical users.

Requirements are a Google account for Drive access and OpenAI for vision/embedding models. Customize by swapping models, e.g., Google's Vertex AI for better multimodal embeddings, or integrating different storage/vector services for scalability.