I’m writing to express my extreme frustration and disappointment with the current behavior of ChatGPT, specifically regarding image analysis. I’ve used this tool extensively in the past, and one of the key features I relied on was its ability to accurately interpret GIMP screenshots and extract pixel-perfect coordinates from clearly defined rectangle selections. That used to work flawlessly.
Now? It doesn’t.
Over the past several attempts, I have sent multiple GIMP screenshots — each showing a crystal-clear rectangular selection — to extract coordinates for FFmpeg’s delogo filter. Despite the images clearly showing the selection and being aligned perfectly with the logo I’m trying to remove, GPT repeatedly returned:
- Incorrect coordinates
- Coordinates outside the video frame bounds
- Or even delogo boxes that miss the mark entirely.
I had to repeat myself four times, each time providing more visual and verbal detail, including failed output screenshots — and even then it still failed. This is not a case of vague prompts. I did the work. I gave the tool everything it needed, just like I always have. But something has changed for the worse.
I don’t know what you’ve altered in the back end — but this regression is unacceptable. I relied on this level of precision for my professional video archival workflow. What used to be done in one step now takes four corrections, and even then I’m still having to finish the job myself.
This tool used to behave like a professional assistant. Now it’s like I’m dealing with someone who can’t even read a ruler on a screen. I am furious and deeply disappointed. If this isn’t resolved, I will seriously reconsider my use of ChatGPT for technical and archival work moving forward. See screenshot and command vs actual result It gave me delogo=x=585:y=400:w=120:h=40:show=0
running it
This used to work something has changed and it’s broken this..
