Incorrect analysis of GPT images in Make.com when using Telegram

How To
, ,
1

Hello, community!

I’m experiencing an issue while setting up an automation in Make.com for image analysis using the GPT module. My automation is structured as follows:

  1. First Module — Telegram → Watch Updates:
  • Receives a user message, which may include an attached image in JPG format.
  1. Second Module — Telegram → Download a File:
  • Downloads the attached image.
  1. Third Module — OpenAI GPT → Analyze Image:
  • Sends the downloaded image to GPT for analysis.

Automation Goal: The user sends an image along with a text query (e.g., “What is shown in the photo?” or “Translate the text in the instruction”). GPT should analyze the content of the image and provide a relevant response.

Problem:

  1. The GPT → Analyze Image module provides incorrect and irrelevant responses:
  • For example, instead of accurately describing the image, GPT gives an incorrect interpretation.
  • When the image contains text, GPT fails to extract it correctly.
  1. However, when the same image is uploaded to GPT via the web interface, it is processed correctly, and the responses are accurate.

My Actions:

  • I confirmed that the file is downloaded and passed correctly to the GPT module (the JPG format is preserved).
  • I tested the same images via the web interface, where GPT produced correct results.

Questions:

  1. Why does the GPT module in Make.com process the image incorrectly, while the web interface provides accurate results?
  2. What steps can I take to improve the accuracy of image analysis in Make.com?
  3. Are there specific recommendations for passing images or parameters that could affect the GPT module’s performance?

Thank you in advance for your assistance!

2

Hi @Nastya and welcome to the Make Community!

In this thread of the OpenAI forum:

There is an example of using the vision API. I used make with the URL given in that example and the description I received was correct.

If you do the same with your module (use the Wikimedia URL), do you get the appropriate response? If so, it might be a problem with the URL ChatGPT is getting. If you don’t get the correct response, then there is likely a problem with the way you call the API.

L