'gpt-4-vision-preview' always starting it's response with the same word

Hi, I’ve tried changing various different ‘temperature’ values but no matter what I do the vision api is always starting the description of my images with the same word - “Nestled”. The rest of the response is different every time.

Does anyone know if there is another parameter I can change to stop this?

Thanks in advance.

I’m no expert on gpt-4-vision by any means but I’ve experimented with it a lot using the API, through scripts I’ve made and apps/UIs other people made for it, and I’ve never had the issue you’re describing.

Is there some kind of system prompt override in the app you’re using? I’m pretty sure that gpt-4-vision does some system prompt behind-the-scenes trickery (telling the AI that it can, in fact, process images), so my guess is that something is interfering with that?