I’ve been trying to make a centaur since Chat GPT let me do AI images and I learned that you can use :: to make things merge, and it worked better. I got one that looked right, but if I tried to make sure it wore clothing it got even worse…
Attempts forcing clothing:
Attempts when I gave up on forcing clothing, it got it right once:
The one that worked:
Hmm. Stable Diffusion seems to want to separate the human and horse too.
Weird right?
I’ve been able to make every other fantasy creature I’ve tried but this one. Makes me wonder why?
If you figure that it’s been trained on very few centaurs and very many images of horses and of people – and of people riding horses – my guess is that it’ll try to make something that looks like what it’s been trained on.
Makes sense. But it can do minotaurs with no problem which I imagine would have the same issue!?
Many times fewer images of bull heads, so a less engrained view of what they should look like.
Yeah could be.
It made a faun?
Haven’t tried. Minotaurs was the closest thing why it did them just fine.
SD (fullyREALXL_v30ForREAL) seems to want to draw one from A Midsummer’s Night Dream with full stage makeup on :/
Probably because 90+% of images it’s been trained on are people that don’t have horse bodies.
Basically it’s over-trained on pictures of people.
Yeah that makes sense.