Lol 😀

The image collection they use to train the AI must be bonkers. OpenAI on the other hand is much better at interpreting what the user wants and delivering results with higher fidelity - closer to what the description says. For example today I tried some stuff with angel wings with different colors and often the NightCafé algo would just ignore that part, or color the background in that tone instead.

I would like to send some back and have the AI expand the image to get the clipped features, but I'm not sure that is possible. Something to ask around.