The images on the site aren’t very good (typical low-detail airbrushed-look) nor are they generated very fast. See the examples here (mostly SFW) on what you can actually do, it takes about 15sec per image on a mid range gaming PC.

That said, one big limit of current AI models remains: It’s always images of a single subject, it can’t do multiple subjects or complex interaction. Also facial expressions still always look quite bland. It can be worked around with inpainting and stuff, but plain text prompts have a hard time generating interesting images.