n8n workflow template · 13 nodes · Google Drive
Generating Image Embeddings via Textual Summarisation is a free n8n workflow template that starts when you run it manually in the n8n editor and works with Google Drive and OpenAI. It has 13 nodes, 7 of them AI nodes (Default Data Loader, Recursive Character Text Splitter, OpenAI Embeddings and OpenAI). The notes inside the workflow say: “Multimodal embedding is not designed analyze medical images for diagnostic features or disease patterns. Please do not use Multimodal embedding for medical purposes.” To run it, you add your own credentials for Google Drive OAuth2 and OpenAI API.
The text below comes from the sticky notes inside the workflow. We removed links, code samples and anything that looks like a key or token.
Multimodal embedding is not designed analyze medical images for diagnostic features or disease patterns. Please do not use Multimodal embedding for medical purposes.
In this demo, we just need an image file. We'll pull an image from google drive but you can use all input trigger or source you prefer.
Read more about working with images in n8n
There are a myriad of image embedding techniques some which involve specialised models and some which do a simplified image-to-text representation. In this demo, we'll use the simplified text representation methods: collecting color channel information and using Multimodal LLMs to produce keywords for the image. Together, these will form the document we'll embed to represent our image for search.
Note, OpenAI Image models work best when image is resized to 512x512.
It is important to define your metadata for later filtering and retrieval purposes.
Read more about vector stores
Once our document is ready, we can just insert into any vector store to make it ready for searching. When searching, be sure to defined the same vector store index used here! Note: Metadata is defined in the document loader which must be mapped manually.
Read more about vector stores
Here's a quick test to use a simple text prompt to search for the image. Next step would be to implement image-to-image search by using the "Embedding Doc" to search rather to store in the vector database.
The text below comes from the sticky notes inside the workflow. We removed links, code samples and anything that looks like a key or token.
Join the Discord or ask in the Forum!
The 13 nodes in file order (sticky notes excluded). Branches and connections determine execution order in n8n:
Configure your own credentials after importing. Create each credential in n8n, then select it in the node.
Do you need somewhere to run it? Read our guide to self-hosted n8n.
We can adapt this workflow to your apps, data and process, then deploy and support it.
Contact Us