Subject: Hello OpenAI team,
I’ve noticed a significant difference in image generation quality between ChatGPT (GPT-4o/GPT-5 web interface) and the Image API (gpt-image-1).
When generating caricatures or stylized portraits:
-
In ChatGPT, the results preserve facial identity extremely well, producing high-quality, recognizable faces (almost production-ready).
-
In the Image API, using the exact same prompt and photo, the results are noticeably worse: faces are distorted, identity retention is poor, and overall quality is lower.
This is frustrating, because as API users we are paying for image generation, but we don’t get the same quality as in ChatGPT.
My questions:
-
Are ChatGPT and the Image API currently using different image models (or different pipelines)?
-
Is there a way to enable the same face identity preservation in the API as in ChatGPT?
-
Is OpenAI planning to release the same improved image model that is already available in ChatGPT to the API (
gpt-image-1or a newer version)?
This would be extremely valuable for developers who want to deliver the same user experience that OpenAI already provides in ChatGPT.
Original:
promt:
Transform this photo into an ultra-realistic, highly detailed 3D caricature with a lifelike, expressive finish. Preserve and emphasize the subject’s unique facial structure, skin texture, hairstyle, clothing, and posture with extreme precision. Ensure the final result remains instantly recognizable and faithful to the original image, maintaining the distinctive character traits and mood. Highlight exaggerated expressions and a stylized, oversized head while using micro skin details (like pores, wrinkles, and facial lines), and lifelike fabric textures. The style should resemble high-end animated film characters.
