I have been playing around with it using ROCM+6900XT, makes a good alternative to DALLE. They have different strengths, DALLE seems better at lighting instructions and cityscapes, but Stable Diffusion is better at sketches.
Also, you can fine tune it on whatever you want which is awesome.
One interesting effect I have noticed on myself though is that after staring at DALLE or Stable Diffusion generated images for a long time then viewing "real" media, I get the same sense of wrongness that the output is not quite right for awhile, like my brain has been tweaking its processing to prefer the AI art as the ground truth!
That's funny, for me dalle2 is in practice miles ahead on pencil sketches, but stable diffusion is cool because the parameters can be customized, which helps with many phrases. Also, you can just leave it running and producing images for an hour.
Also, there's no content filtering, but I don't recommend playing around with that if you're sensitive. The lifeless husks and various mixes of body parts I got when playing around with it with fairly benign phrases could very well be used for a horror movie.
It might be that I haven't yet found the right phrase for stable diffusion for pencil sketches though, as for dalle2 it's just "<describe what you want>, artstation, pencil sketch, 4k" to generate consistently great pictures.
4chan is having a field day with AI generated porn of celebrities (often with ridiculous prompts) and selecting the most unsettling. One for Billie Ellish looks like some kind of orphaned shoggoth/succubus hybrid just made its first attempt at luring in someone for a meal: "You like human females, yes?" Cataract eyes, aggressive lobotomy mouth, it forgot to pay attention to shoulders and didn't know spines existed or what they were for. Or a second attempt, this time at Bjork, suggesting some kind of lost hominid which consumed only melons in a predator-rich environment.
Also, you can fine tune it on whatever you want which is awesome.
One interesting effect I have noticed on myself though is that after staring at DALLE or Stable Diffusion generated images for a long time then viewing "real" media, I get the same sense of wrongness that the output is not quite right for awhile, like my brain has been tweaking its processing to prefer the AI art as the ground truth!