I'm not sure what algorithms the rating system uses, but I think we can agree it needs review. Pictures get analysed with an absurd rating. Or... they disappear into the ether, re-appearing a day later with an absurd rating.

PG-13? Either the algorithm is over-tuned, or this is the sexiest living room ever.

The funny thing about the above image, is that the AI tagged it fairly accurately

NO HUMANS - important tag it produced here. How does it get a PG-13 rating from this?

Perhaps misrating PG to PG-13 isn't that bad in effect, as the picture should still show up in the SFW site. But getting a PG-13 picture rated as "R" means the picture is confined to the NSFW site, shutting it off from a wide audience during the crucial first moments of publishing. It also means that frustratingly, yellow buzz is needed to unlock the picture from the generator, even if it's not really an R+ image!

It's not really known what tips a picture from PG-13 over to R+. Is it the skirt/pants being too short by some arbitrary length? Back shots seem to trigger the algorithm more than frontal shots - because there is more focus on the backside?. Large breast size seem to trigger the algorithm more as well, even if the character's outfit is exactly the same. But then on the flip side, having too small breast size might lead to others thinking the picture is a ToS violation.

Maybe someone with a lot of time to burn and the resources to spend can try this experiment. Generate a clothed female, and slowly adjust just one thing at a time - e.g. bust size, skirt length, amount of cleavage seen, while keeping everything else the same. Maybe you'll be able to figure out the exact tipping points. Good luck!

Edit:

🤦‍♂️

Had to change the cover picture, again!