Some things I was able to put my finger on after looking at a bunch of the images in the feed:

• It doesn’t do skin well, treating it more like a smooth plastic than a surface with pores, wrinkles, and fine hairs.
• It doesn’t understand lighting, so shadows don’t agree with highlights or even each other.
• It doesn’t understand anatomy. A lot of the images were fine in this regard but others had misplaced muscles, bones, and impossible limb positioning/truncation.
• It has no idea how to draw vaginas. Nipples are also not well understood, though it does better on average with those. They still look more like a plastic than skin, but most of them were passable at least, while I didn’t see a single vagina that looked even close to right.

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.