Hello,
I’ve created an OCR scenario that takes PNG files from FTP and processes them using GPT-Vision to get indexes and barcodes from it.
When I upload files directly into ChatGPT, it performs OCR perfectly every time. However, when I make API calls through the Vision Module (gpt-4o) in Make.com, the results are consistently poor. What could be wrong? Does it degrade the image quality, causing the model to generate inaccurate results?