GPT-4V "detail" parameter doesn't work

API Bugs
1

Hi!
I have a problem with detail parameter. No matter whether I set it to low or high, cost of image in base64 format passed to gpt-4-vision-preview is 765 tokens (as for 1024*1024 in high detail) in contrast to 85 for low detail image according to documentation.
Does somebody know what I can do about it?

Here is example of request body:

payload = { "model": "gpt-4-vision-preview", "messages": [ { "role": "user", "content": [ {"type": "text", "text": "Describe the images one by one."}, *map(lambda x: { "type": "image_url", "image_url": { "url": f"data:image/jpeg;base64,{x}", "detail": "low", }, }, images[:70]) ] } ], "max_tokens": 1000 }

2

Try Reducing the max_tokens then it will generate less tokens = less cost. I hope these below resource will help.

 {
            "type": "image_url",
            "image_url": {
              "url": "https://commons.wikimedia.org/wiki/Main_Page#/media/File:Die_Woche_Der_Tag_von_Potsdam_cover.jpg",
              "detail": "low"
            }
          }
        ]
      }
    ],
    "max_tokens": 300

https://platform.openai.com/docs/guides/vision/low-or-high-fidelity-image-understanding

3

Hi, I have noticed this was working last week but it doesn’t appear to be working now. Did you get this fixed or is this a bug?