I’ve also found that when you ask it to show someone “drawing a bow” they end up with a strange mix of a pencil that goes up to their cheek, and another weird stick going perpendicular to whatever they’re doing, and other weird bow-related artifacts. It’s pretty funny. (At least with SD 1.5)
Synonyms are stable diffusion’s enemy. OpenAI is way ahead on that front, but there’s a good reason for that: They invented CLIP, so they probably have better checkpoints than anyone else.
I’ve also found that when you ask it to show someone “drawing a bow” they end up with a strange mix of a pencil that goes up to their cheek, and another weird stick going perpendicular to whatever they’re doing, and other weird bow-related artifacts. It’s pretty funny. (At least with SD 1.5)
“drawing a bow” →
Synonyms are stable diffusion’s enemy. OpenAI is way ahead on that front, but there’s a good reason for that: They invented CLIP, so they probably have better checkpoints than anyone else.