GPT-4o Occasionally Refuses to Analyze Images?

[lavish coral]

Hello everyone,

I am currently using GPT-4o's vision capabilities in one of my projects. Put simply, i scrape images off of the web, resize/process them and then send them over to OpenAI's API which analyze said images and generates tags, based on their contents.

However, upon testing this functionality i find that on occasion, GPT-4o refuses to analyze images of any nature. Instead, it provides me with a message that usually looks something like this:

"I don't have the capability to view, recognize or analyze images directly. Please provide a description of the image, and I can help you based on that information."

This only occurs in about 20-25% of instances. The rest of the time, it works perfectly as intended. This is not a fatal error, but more of a slight nuisance/annoyance.

I was wondering if anyone else has experienced similar issues with GPT-Vision? Specifically with the new GPT-4o model? Perhaps my prompts are off?

[remote gull]

I have seen this as well, I would try this as silly as it sounds

in your system, put something along the lines of "Image Recognition: Enabled"

[lavish coral]

I'll give it a try. I tried asking ChatGPT, what i could do to fix this (y'know ask the AI what the AI wants), and it told me told remind the model, in my prompt, that it can analyze images. So gaslighting my model to analyze my images, helped a tiny bit. I'll report beck on whether your suggestion yields better results 😅

[lavish coral]

Okay update:
Your suggestion actually helped a bit. The model still refuses to analyze pictures at times. Sometimes even without prompting a response entirely. But it works for now, my hope is that OpenAI will update the model in the future, to avoid scenarios like this

[lavish coral]

Looks promising. However, my use case depends on OpenAI's vision capabilities. We already have an established implementation, featuring OpenAI's other capabilites