Built an extraction tool for my own projects, then realised it describes product images too

Built with n8n
, , ,
1

:waving_hand: Hey n8n Community,

Bit of background first. I originally built this extraction setup to serve my own automation projects, mostly document classification and pulling structured data out of PDFs and invoices. It worked so well for me internally that I decided to open it up so anyone can use it.

The new use case

In my last project I needed product photos described so I could feed that into content generation, and I almost reached for a separate vision model. Turns out the Extractor handles images really well, which is the moment it clicked that this is more than a classic data-extraction tool.

How it works

Instead of pointing a pipeline at a document, you point it at the image and define the fields as questions about what’s visible: dominant colour, visible features, material, the type of shot. It comes back as structured fields, exactly like a document extraction, so the result drops straight into the rest of the workflow with no paragraph to parse and clean up.

The workflow

I recorded a short video walking through it, attached to this post. You can also see the actual workflow I built around this case here: n8n-workflows/easybits-product-content-creation-workflow at e3103344d9b3358402dc38a3a862d510bb4e7c5e · felix-sattler-easybits/n8n-workflows · GitHub

Setup

  • n8n Cloud: it’s a verified node, just search easybits Extractor in the node panel. Nothing to install.

  • Self-hosted: Settings → Community Nodes → Install → @easybits/n8n-nodes-extractor.

Then create a pipeline, define your fields, and point it at an image instead of a doc. There’s a free plan with 50 API requests a month included, which is plenty to test the image use case end to end.

Anyone else using extraction tools for image understanding rather than a dedicated vision model? Curious what you’re pointing them at.

Best,
Felix

2

Welcome @easybits!

The structured-output approach is what makes this immediately useful in automation - instead of parsing free-form AI text downstream, you get clean fields that drop straight into a Spreadsheet or Airtable node. The product image description use case is a good one; another solid fit would be warranty/spec sheets where you need to extract specific attributes from a mix of PDFs and product photos in the same pipeline.

3

Hey @nguyenthieutoan, 100% agreed! I think generating product image descriptions is one of those use cases where AI can provide immediate value. Writing those descriptions manually is not only time-consuming, but also highly repetitive, making it a great candidate for automation.

The warranty and spec sheet use case sounds really interesting as well! Since I’m already working with product specifications in the current setup, I’ll definitely take a closer look at that idea.

I’m curious though, are the warranty PDFs usually multi-page documents with lots of boilerplate and unnecessary information? If so, that sounds like another great use case for targeted data extraction.