This problem has been discussed several times in this forum, but mostly based on subjective feelings. This time, I will show you a result that you can never get from the Dall-E-3 API, but ChatGPT provided you with.
Look at the images generated by the same prompt
Yes the first one is generated by ChatGPT, the second one is by the API, the prompt is ‘A single 3D cartoon-style character inspired by Kenny McCormick from South Park, featuring a young boy with a round face, wearing an orange winter outfit with a hood that covers most of his head, standing confidently in a snowy mountain environment with falling snowflakes and a whimsical, cheerful winter atmosphere. The rendering includes smooth 3D textures, vibrant colors, and a playful design.’ (Even this was written by ChatGPT)
I’ve tried many differenct characters and even the group photo of different characters, ChatGPT can always give you the accurate result, but never success with the API.
I think ChatGPT is using an internal version of Dall-E-3, with much more loose content filter and trained with much more data.