So here we Queen Serenity, the upgrade model, a bustier more mature and womanly version of her daughter.

It was a pain to make. The fanart is few and far in between, also for some reason most of it has huge horrible looking watermarks. So screencapping i went then water mark cleaning.

Then everything went fine right? Right?! Obviously not because that never happens. Turns out 99% of the images had her crescent_facial_mark making it impossible to generate without it. So back to cleaning i went this time creating an artificial subset of images without her mark for contrast so SD could learn the difference. All in all it was a pain.

Works fine with <lora:QueenSerenityV3:0.7> queen_serenity,

Anyway only one outfit and her toggleable mark:

• white_strapless_dress_cleavage_crescent_brooch_white_bowtie_white_back_bow: Her og dress

• crescent_facial_mark: she forehead mark, as expected it is iffy due to SD limitations as it is more complex that usagi's (looks like this ...🌙... 3 dots and the crescent).

Anyway had to explicitly train the crescent so it could be deactivated.

I will add tag summary only if people want to experiment. Can't add the dataset as i did a screencaping run for OG, R and crystal.