How Does GPT-4V Image Resizing Impact API Costs and Technical Drawing Detail?

Resizing is Expected Behavior

• Normal Operation:
  The API automatically resizes images before processing. This is designed to optimize performance and ensure token counts remain manageable. For instance, even if you submit a large technical drawing, it will be resized to fit within internal constraints. This is why the cost estimator indicates a resized resolution.

• Effect on Descriptions:
  Resizing might diminish the legibility of very small details such as annotations if they become too small. The documentation advises ensuring that important content (like text) is large enough so that, even after resizing, it remains clear. For technical drawings with annotations, you may need to ensure:

  • The text is sufficiently enlarged.
  • Critical details are not cropped out during the resizing process.

Recommendations

• Enhance Detail if Needed:
  If the annotations or small texts are crucial, consider adjusting the image before sending it to the API so that these elements appear clearly even after resizing. Sometimes, enhancing the legibility of technical drawings beforehand can help mitigate any potential loss of detail.

• Detail Setting:
  Depending on the use case, you might evaluate whether using a higher detail setting could be beneficial. However, this will increase the token cost.

By taking these steps, you can effectively balance cost and quality for describing technical drawings using GPT-4V.