Ask HN: GPT 4 Vision can't tell the difference between fonts. Am I asking wrong?

I'm trying to identify whether an image of text has been doctored, sloppily, by using the wrong font.

GPT 4 Vision API suggests that all the fonts in the image are the same "Calibri" font, even when I change one word to an obvious joke font.

Is there an approach or trick for getting GPT 4 to identify that one of the words in an image uses an entirely different font?

Is there a simpler or cheaper approach than using GPT 4?

2 comments

[ 2.9 ms ] story [ 17.7 ms ] thread

sharemywin 2y ago ↗

might check these guys out. I never used the api but I think this was the site I uploaded an image to

https://www.whatfontis.com/API-identify-fonts-from-image.htm...

verdverm 2y ago ↗

> Is there a simpler or cheaper approach than using GPT 4?

I'm not sure this is something GPT-4 can do reliably, if at all. You can definitely find better models with higher performance that will be much much cheaper.

Is there something available out-of-the-box? Probably not, you are asking a rather unique question. You should be able to fine-tune or train one up though, given sufficient training data. You probably want to start from an OCR model.