>>11336077
the basic idea is
>there are different models for generating images
>each model has been trained using a big set of images
>the images are tagged with what's in the image, so the model associates a word with certain concepts
>make prompts with words that the model is trained on

for most muscle artists though, they probably aren't popular enough for ai to have data on them
but people can somewhat easily make Loras, which is like a little add-on for models to generate whatever specific concept you want