AI Generated Images

i find ai is not the best at understanding actual creativity. It has a tendency to see the unusual part of what you are saying and think "that's abnormal, surely that's not what is meant, i better do the normal way". I find in SD1.5, since it's old, it at least SEEMS creative because of the chaos. But I often go to SDXL or SD3.5 with an extracreative idea I want to be really specifically done perfectly and, instead, it makes a 'perfect' image not of what my idea was but of the nearest human commoner concept a noncreative boring person with no creativity would have. As ai modernizes, it seems better at images that don't think outside the box, and converts prompts outside the box to being within.

Also I hope my current view is incorrect and I HEAR midjourney (which i haven't used) is good at prompt specificity

Some generators censor certain words. I can imagine that it's possible that "shit" or similar is in the list.

EDIT: Some experimentation with a model (realmixXL) that clearly doesn't in ComfyUI: I can get a log cabin made of shit, especially if I crank down the prompt term weight, but it doesn't seem to generate logs made of straight feces. I'd guess that maybe there aren't enough photographs of straight feces to be able to pick up on that as being an option.

e.g.

A (log cabin:.7) made of poop in the woods.

Other substances clearly work:

A log cabin made of ice in the woods.