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 &lt;lora:QueenSerenityV3:0.7&gt; queen_serenity,Anyway only one outfit and her toggleable mark:<ul><li>white_strapless_dress_cleavage_crescent_brooch_white_bowtie_white_back_bow: Her og dress</li><li>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).</li></ul>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.

Queen Serenity - Sailor Moon

QueenSerenityV5.zip

QueenSerenityV3.safetensors

v1.0

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 &lt;lora:QueenSerenityV5:0.7&gt; queen_serenity,Anyway only one outfit and her toggleable mark:<ul><li>white_strapless_dress_cleavage_crescent_brooch_white_bowtie_white_back_bow: Her og dress</li><li>crescent_facial_mark: her 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).</li></ul>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.

Queen Serenity (DORA version) - Sailor Moon

queenSerenityDORAVersion_v10_trainingData.zip

queenSerenityDORAVersion_v10.safetensors

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 &lt;lora:QueenSerenityV4:0.7&gt; queen_serenity,Anyway only one outfit and her toggleable mark:<ul><li>white_strapless_dress_cleavage_crescent_brooch_white_bowtie_white_back_bow: Her og dress</li><li>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).</li></ul>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.

Files with SHA256: c647cdf0ba70866340d16c302a033cc7e36e6bf9e25be633354b6e43987215eb