Sign In

Usagi Tsukino (fanart) - Sailor Moon

539
3.2k
1.2k
28
Verified:
SafeTensor
Type
LoRA
Stats
1,881
1,197
3
Reviews
Published
Aug 9, 2023
Base Model
SD 1.5
Training
Epochs: 8
Usage Tips
Clip Skip: 2
Trigger Words
SMUsagiTsukino
ClassicSM_Blue_skirt_White_gloves_Red_boots_Red_choker_Red_bowtie_Red_back_bow_Circlet_Red_bun_cover
EternalSM_Yellow_layered_skirt_Pink_puffy_sleeves_White_gloves_White_boots_Wing_brooch_Red_choker_Red_ribbon_Wings_Hairpin_Red_bun_cover_Golden_crescent_forehead_mark
SuperSM_White_skirt_White_gloves_Red_boots_Yellow_choker_Red_bowtie_White_back_bow_Crescent_circlet_Hairpin_Red_bun_cover
Princess_white_strapless_dress_huge_back_bow_Golden_crescent_forehead_mark
school_uniform_blue_serafuku_blue_back_bow_red_bowtie
Training Images
Download
Hash
AutoV2
50F55D5A98
default creator card background decoration
knxo's Avatar
knxo

For gentle people of culture only! This was trained almost exclusively on NSFW material.

V2: I spent a whole week poking this and spent all the tricks in my sleeves and some new ones. V2 or rather V15 is finally here. At this point I an not sure if it is better than the old one, but it has better concept separation on her tagged outfits, though anything dress seems a bit corrupted by her princess gown. It should also be less prone to deepfrying. I'll probably bake the LOCON version tonight so you can expect it tomorrow.

Works fine with: <lora:SMUsagiTsukinoV15:0.7> smusagitsukino, <outfit>

I managed to more or less isolate the following outfits, I had to be stupidly specific to maximize separation as i kept getting a virtual soup:

  • ClassicSM_Blue_skirt_White_gloves_Red_boots_Red_choker_Red_bowtie_Red_back_bow_Circlet_Red_bun_cover

  • EternalSM_Yellow_layered_skirt_Pink_puffy_sleeves_White_gloves_White_boots_Wing_brooch_Red_choker_Red_ribbon_Wings_Hairpin_Red_bun_cover_Golden_crescent_forehead_mark

  • SuperSM_White_skirt_White_gloves_Red_boots_Yellow_choker_Red_bowtie_White_back_bow_Crescent_circlet_Hairpin_Red_bun_cover

  • Princess_white_strapless_dress_huge_back_bow_Golden_crescent_forehead_mark

  • school_uniform_blue_serafuku_blue_back_bow_red_bowtie

  • golden_crescent_forehead_mark: use this in the negatives to get rid of it or boost it.

If It is being uncooperative try to boost the concepts you want, I managed to make it cooperate and finally produce a red cheongsam by using (red China dress). Or you could lower the weight a bit.

Lot's of tagged stuff inside so download the dataset with the tags summary and experiment.

V1

Here it is finally. I had my doubts on making it correctly. If by any chance you want some lolishit then begone, this was trained mostly on Kuroinu Juu and glamourworks.

There are several good enough Sailor Moon Loras which is why I kept postponing this, also the NAI model family do a good job by themselves in a pinch. But I wanted my own SM Lora with Black jack and hookers!

I managed to more or less isolate the following outfits

  • SMEternalOutfit (yellow and black skirt might require puffy_sleeves as the Kuroinu variant often is sleeveless )

  • SMStandardOutfit(Normal outfit, for some reason has the highest variability despite the largely homogeneous dataset. so it might need some retries.)

  • SMPrincessOutfit (came out nicely it is her white and gold dress. )

  • SMSuperOutfit (somewhat iffy white blue and yellow skirt has some bleedover but works fine enough)

  • SMSchoolUniform (Normal school uniform which resembles her standard outfit.)

  • SMPartialOutfit (I dropped in here all partial outfits and bikini versions, sadly due to contamination it actually mostly resembles her standard outfit. You should have an easier time removing parts of it (if you only want her leotard or her torn uniform).

There seems to be some contamination between outfits, I might do a V4 Later. Yup this is actually V3. I had forgotten how much contamination invoking sailor moon in the triggers can cause. So V1 was a mess, V2 lacked separation in the outfits in the dataset and made me unhappy and this is V3, good enough but I get the feeling i might have gotten further separation(or less bleeding) if I made the outfit triggers more unique.

Anyway I'll add the tags summary in the dataset if people want to experiment. This was trained basically on porn doujins so be warned.