I once used the term “beefy” (as in big or muscular) with good results. There actually was “beef” in the output as well but it just didn’t look as good when I stopped using that word.
So from what I’ve noticed, AI tends to make women impossibly skinny by default. Bing really didn’t like talking about women’s bodies, at least in the Dalle2 iteration. Athletic often resulted in just as skinny women. Fat got flagged. Hourglass figure got flagged. Normal body proportions didn’t do anything. Anatomically correct is a mistake.
Bing did like buxom, which it understood to mean morbidly obese, and “kinda-” and “semi-” are modifiers it accepted. So if I’m trying to generate a picture featuring a humanly possible woman, I’ll use the phrase “kinda-buxom” or “semi-buxom.”
I wish I understood what punctuation and symbols do in a Dalle prompt.
I’m not 100% sure how much punctuation matters outside of the general prompt formatting. You seem to be able to get away with spelling errors so I can only assume there’s some sort of error correction/interpretation layer in play though I can’t say how robust it is.
I once used the term “beefy” (as in big or muscular) with good results. There actually was “beef” in the output as well but it just didn’t look as good when I stopped using that word.
So from what I’ve noticed, AI tends to make women impossibly skinny by default. Bing really didn’t like talking about women’s bodies, at least in the Dalle2 iteration. Athletic often resulted in just as skinny women. Fat got flagged. Hourglass figure got flagged. Normal body proportions didn’t do anything. Anatomically correct is a mistake.
Bing did like buxom, which it understood to mean morbidly obese, and “kinda-” and “semi-” are modifiers it accepted. So if I’m trying to generate a picture featuring a humanly possible woman, I’ll use the phrase “kinda-buxom” or “semi-buxom.”
I wish I understood what punctuation and symbols do in a Dalle prompt.
I’m not 100% sure how much punctuation matters outside of the general prompt formatting. You seem to be able to get away with spelling errors so I can only assume there’s some sort of error correction/interpretation layer in play though I can’t say how robust it is.
no- is an strange prefix. No-knees means nothing covering the knees. No-torso means topless.
That could be very useful, I’ll definitely have to do some testing with that.