Why do longer prompts fail on Gemini 1.5 Pro via n8n, while short ones work fine?

Questions
1

I’m facing a strange issue while integrating the Gemini 1.5 Pro model for image analysis using n8n.

The flow works perfectly when I send a short prompt (like a single line). But when I use a longer, structured prompt with multiple steps and detailed instructions, the HTTP Request completely fails (not just a bad response — it fails outright).

I’ve already confirmed that:

  • The request is going to the correct endpoint:
    https://generativelanguage.googleapis.com/v1/models/gemini-1.5-pro:generateContent
  • The JSON format is valid:
    contents → role: "user" → parts → text
  • I’m using POST with Content-Type: application/json
  • The API key is working fine for other requests.

Has anyone run into this? Is there a hidden payload size limit? Or could the model be rejecting the input for another reason?

Any insights would be greatly appreciated!

image
image948×249 33.3 KB

2

check the token limits. each LLM has different limits on input and output tokens