Skip to main content

Had a quick question, when using "Analyze image content with vision”

I have some images come in where it will have text like Date: 04-01-2024 but the text itself might be an unclear scan. So GPT usually gives me the wrong answer when I tell it to extract dates i..e turns 4 into 8 at times, 7 into 2 etc. which getting all other digits right 

I can’t really control the source file quality as it is what I need to work with. But interestingly, if I upload the image and give the same prompt on my consumer version of GPT 4o, it manages to extract the correct data points. 

Anyone facing the same issue?? 

Hi @Imrans23, welcome to the Community! 🎉

I did some checking on this end but couldn’t find any bug reports mentioning similar cases of Analyze Image Content With Vision actions incorrectly interpreting numbers. That action uses GPT-4o so I’d have thought it should be capable of interpreting the images in the same way as they are when uploaded directly to ChatGPT.

It could be that this is a bug of some kind 🤔 I’d suggest reaching out to our Support team directly about this here. They’ll be able to dig into the logs for the Zap to confirm whether this is a bug and can get a bug report opened so our engineers can then work on getting this fixed. 

Keen to ensure this gets sorted so please keep us updated on how you get on with them!