I’m finding that DallE3 is very good at ignoring negative requests in prompts, that is, to specifically not include something in the image. It just takes the key word and includes it anyway. Has anyone else found this?
For example, if I offer a detailed prompt to create an image of “a half-orc with pointed ears and longer tusk like teeth, without head horns, wearing leather armour, attacking with a large axe, silhouetted against a swirl of raging yellow, red and orange, make the background transparent”
It gets pretty much everything but absolutely refuses to not include horns on the head. It sometimes doesn’t produce a transparent background either.
Has anyone found a way to word a prompt to specifically not include something? This feels like a gap in the model so far.