Approximate summary of my use of generative art AI so far:

Whenever I try to use it with the goal of making something pretty that dazzles me, I am very impressed on the first try.

Whenever I try to use it to draw something specific thing that I have a need for, I get very frustrated.

This was also my experience with using AI for text and code tasks a year ago, and since then I don't feel that way; I am not sure what portion of that is due to the AI improving and what portion is due to my mind adapting and instinctively not even trying to get the AI to do things that it is bad at doing.

learning to prompt is no different than learning a programming language - more accessible and therefore more difficult to master

I've definitely avoided 0-shot experiments other than basic gimmicks. in many cases it would reasonable to assume that the task would be considered as multiple procedures w/ multiple perspectives in a human context. controlnet can be slightly useful, but it remains to be seen if it can be further improved/downscaled.