Hi! I want to be using gpt-image-1 model through api, and i expect at least 50 images to be generated for my app’s users per month per user, a 1024 x 1536px portrait image. given this if i use the high quality parameter it seems the estimated cost for 50 images might reach ~$13.5 at $0.27 per image which is huge, so in this case if i could use medium quality 1024 x 1536px portrait image it would significantly reduce the cost, but the images need to contain mistakes free text which i could see from sora generation that it could handle text rendering really good without mistakes if the text are kept minimal, but i am unsure of the results of medium quality image generation, has anyone else used medium quality parameter and how have the results been particularly the quality of text rendering.
All of these have a bit of reality-breaking inconsistency (reproducing the underscore used in the prompt is the same as previous use.)
So medium, also below, can be pretty good, it just has a symptom of a bit cheaper upscale that does a bit of swirly squiggle, on perhaps a lower resolution AI output. A “still AI” look. It is not the amount of text that will be the biggest cause of this artifact.
FYI: another provider’s model seems to have a different idea about the sign, and produces reality and sharp letters that are not an immersive part of the photograph, but can’t remake a whole web page screenshot in cartoon style.
