#PSA: "detail" parameter for gpt-4-vision API
1 messages · Page 1 of 1 (latest)
bumping this
ty
What's the default?
auto
i think its at high as default? I ran it with some HD images and i could tell it was set to high because of the token usage
nice, these engineers and researchers at openai can create the most magical stuff but their documentation isn't that clear lol
was so confused on how to pass the detail parameter
Might just be forcing high for those image inputs
I'm sure it's a hectic time! They'll definitely polish the docs
What exactly is the difference in functionality? In other words - which use-cases need 'high' resolution and which use-cases need 'low'?
low resolution costs less
high resolution theoretically should always be better than low resolution since you're giving more data for gpt to work with
but low resolution costs like 1/4th high resolution depending on image size
Need not be. Really depends on the distribution on which the model has been trained on
i mean sure