It’s trial and error, with a bit of educated terms mixed in. The AI works better with furry than fluff for example, and if you say pony instead of baby horse it adds in a slight chance to toss some MLP dna into the mixture, hence the randomly pastel colored abominations.
I still have to figure out a more reliable way to get this AI to process stuff like eating spagetti or using a litterbox. If you just say spagetti it usually uses raw noddle images, but pasta seems to imply cooked dishes. Also if I ever get access to dall-e 2, the image quality and accuracy will be scary good by comparison. This is a casio wristwatch compared to dall-e 2’s smartphone level sophistication.
Yeah, every so often searching the same thing over and over strikes gold. You just need it to draw from the right randomly selected google image for reference to get what you’re looking for. In that image’s case, about 3 decades of inbreeding.
say ‘drawn in crayon’ at the end if you want it in crayon style. Or any style you want it in. Minecraft, hr gieger, dali, 3d ps1 graphics, it can render a lot of styles.