Can one train AI for n8n?

Questions
1

Hello, I’m trying to build an AI image analysis tool that writes an estimate, or at least identifies the damaged parts, of a car from a POST request with the VIN and an image of the damage. I vibecoded a site lol that sends JSON of the VIN and binary of the image via POST and then edits the image to put the VIN in the top left corner, and I got big company LLMs such as claude and chatGPT to read the vin in the top left then identify the damage and respond with a JSON format of the damage and the car type my site can read and then show the response in a nice format to the user. But now I want to train my own model - I tried huggingface but I can only get that to work with a chat response for some reason and when I try it with a POST request it says something about conversational only. I dont want my site to be like a chatbot. And all the other tools such as AIcado and unsloth do not work with n8n. does anyone know how to do this? Thanks

2

@riversong87871 I don’t think training a model is necessary. Have you considered using Qwen VL models? They’re amazing at identifying objects and can even provide coordinates. Gemini 2.5 Pro also has impressive vision capabilities

2 Likes
3

I haven’t tried Qwen yet. Can you help me with setting up Hugging Face to work with a POST request? I’ll also take a look at Gemini. Currently, I’m using Opus 4.1 and have experimented with GPT 5

4

I’ve used qwen via openrouter and had it examine images. There’s some free qwen versions on there but they’re slower so I use the paid ones. I use python for the requests.

2 Likes
5

Just tested the gemini 2.5 pro on openrouter in my automation and so far it is really good.

Thanks for the attention

6

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.