Method to set the image processing quality #76
Conversation
OpenAI vision can either use low res or high res. Low res is cheaper and uses only 85 tokens. High res uses 85 tokens + 170 * tileCount, where tileCount is the number of 512x512 tiles the image is split into. High res scales the image down so that the longest edge is <=2000px, and the short edge is <=768px.
|
You can directly call |
Though it's possible to do this, you still need to cast the What would you think about adding a new builder function to include the detail level? I did consider it, but I thought it'd be too bulky to have two more methods with similar names to the existing ones. |
|
I am thinking whether it is necessary to create a |
|
It probably would be beneficial to create a builder class. |
|
Tomorrow, I will add this feature |
OpenAI vision can either use low res or high res. Low res is cheaper and uses only
85 tokens. High res uses85 tokens + 170 * tile_count, where tile_count is the number of 512x512 tiles the image is split into. High res scales the image down so that the longest edge is <=2000px, and the short edge is <=768px.To save on token cost or to improve quality, this change allows the users to choose processing quality.
https://platform.openai.com/docs/guides/vision#low-or-high-fidelity-image-understanding