UNSOLVED: How can I store images in a knowledge base (vector store), and what is the process for retrieving them later?

[Robert_Spears]

I’m working on building a knowledge base using a vector store and would like to store images alongside text data. My main questions are:

1. How can I add images to the vector store?

• Should I convert images into embeddings first, or is there a way to store them directly?

• What tools or integrations (e.g., OpenAI, Pinecone, Weaviate) work best for handling images?

2. How can I retrieve the stored images later?

• If images are embedded as vectors, how can I ensure accurate retrieval?

• Are there specific workflows or nodes in n8n that can help with this process?

Any guidance, best practices, or examples would be greatly appreciated! Thanks in advance!

Here are the answers to the typical asks:

n8n version: 1.122.4 (self hosted)

• Database (default: SQLite): Qdrant Vector, version v1.16.1, in Docker Compose self hosted server

• n8n EXECUTIONS_PROCESS setting (default: own, main): main, trying to avoid sub-processes

• Running n8n via (Docker, npm, n8n cloud, desktop app): Docker Compose self hosted server

• Operating system: Ubuntu 24.04 self hosted server running Ollama, with Internet access for API connections if needed

[Epicop]

Vector stores work with embeddings, not raw images. You need a multimodal embedding model like OpenAI CLIP or similar to convert images into vector representations, then store those embeddings alongside metadata containing the actual image URL or base64 data.

Store images separately in file storage or database, then embed a reference or description. When retrieving, the vector search returns the closest match, and you pull the actual image using the stored reference.

Qdrant supports metadata fields, so store image paths there. Use HTTP Request or custom code to handle multimodal embeddings since most n8n vector nodes focus on text only.

[Robert_Spears]

Can you provide an example of this?

I was able to find all this information already in N8N Docs. Even the N8N Chatbot provided a very similar answer. What I need is a node based example. I’ve built plenty of workflows, so I understand the premise.

[Parintele_Damaskin]

Hey @Robert_Spears !

Don’t mind the first response of thr humanBot lol…

I have followed a tutorial where it was recommended by a staff member, where I used PDF files that contains images and text :

How do I create a workflow that retrieves images from the database when I ask the agent to display them?

Hey @astrumax hope all is well. Welcome to the community. A few things that I want to mention here - storing images in the database is not always the best way to go about it. What you want most of the time is to store them in some sort of storage and have a way to link them with URLs. This way you would have to store the references in the database, instead of images themselves. A solid example of doing it this way could be found here, they are using Supabase for storing files alongside their e…

For your case, where somehow you have 2 objects (text and binary) , it will be relevant to store the images somewhere, and create a rich metadata(URL,description, title, alt etc.. ), and save that in Qdrant …

If you manage to accomplish your case somehow, please let me know as well !

Cheers!

[Robert_Spears]

I was about to leave a snarky response, but that wouldn’t be in the spirit of the community. However, charging $779 for community based help seems a bit steep. Thanks for your input, but correct me if I’m wrong, I thought community based help should be free. If you have the workflow that was described in the video, that would be awesome. I’ve looked over this, and a couple of other forums, and it seems like everyone has this type of documentation locked behind a paywall.

Thank you for the encouragement, and yes I will figure this one out. It just seems I will be one my own for this project.

[Parintele_Damaskin]

@Robert_Spears

thought community based help should be free.

Honestly I don’t understand that statement.

I use the community free version of n8n, with a license so I can get access to some features like folders, DEBUG I editor etc…

Do you a bought license that offers…?

@Robert_Spears

If you have the workflow that was described

Not quietly the same, adapted to my use case(but I will share with you my setup once I get to PC)

@Robert_Spears

looked over this, and a couple of other forums, and it seems like everyone has this type of documentation locked behind a paywall

Yea, most services are “freemium” and some tutorials are I indeed behind a paywall.

But anyway did you checked the free workflows that n8n distribute from some creators (there are free and paid as well).

Cheers!

[Robert_Spears]

I really do appreciate the input. I self host N8N, as well as Ollama, Qdrant and Mongodb. My goal isn’t free (hardware costs real money), my goal is to sandbox my data and keep it private.

I have looked at most of the examples across just about every platform and forum I could find. My frustration grew when all of these free examples did not include any JSON or even a PDF of how the workflow was designed. And the videos never showed any details of the nodes. Then to top it off, a link was posted that required a minimum $100, up to over $1,000 for access. All in community forums. People want to make money, and they should. Promote your website, I’m all for that. Posting a paywall link in a community forum just seems very., very, wrong.