Sign In

Mutsumi Otohime - Love Hina

122
614
7
4
Verified:
SafeTensor
Type
LoRA
Stats
32
0
Reviews
Published
May 19, 2025
Base Model
SD 1.5
Training
Epochs: 8
Usage Tips
Clip Skip: 2
Strength: 0.7
Trigger Words
otohime_mutsumi
green_shirt_wing_collar_long_sleeves_brown_pinafore_dress
Training Images
Download
Hash
AutoV2
0CAAE869DE
default creator card background decoration
knxo's Avatar
knxo

One of the OG "Ara Ara onee-san" type characters. Probably the best of the core cast. The only ones who I can stand from that series are Mutsumi, Haruka, Kanako and Motoko's sister Tsuruko. The rest of the girls just give me varied levels of annoyance.

V2

Holy crap this was an extreme pain. Not many images so i had to do a screencaping run and manga cropping.

Most of her images are with her eyes closed and there are very few repeated images from her outfits. The V1 I did long ago when i was dumb and new so i still tried overbaking it quite a lot to make it spit a couple of poorly represented outfits. The manga to the rescue? Nope! All anime costumes are anime only and manga costumes manga only.

I did learn something! Mutsumi might actually be a christian as she uses a cross keychain in her bag and a crucifix in the neck on one of her wedding scenes.

Regardless after cleaning and suffering turns out Ken Akamatsu did a good job with the anime art it being suspiciously similar to the manga, good right? No! It took the style too strongly.

Regardless I ended making 2 versions a "normal" one and one with Normalization to kill the style a bit.

For the style heavy one(No normalization):

Works fine with <lora:MutsumiOtohimeV3:0.7> otohime_mutsumi, single_braid

For the Normalized one(in V4 the normalization went too far):

Works fine with <lora:MutsumiOtohimeV5:0.7> otohime_mutsumi, single_braid

I gave up on most of the outfits as there are just too few images so we get:

  • green_shirt_wing_collar_long_sleeves_brown_pinafore_dress: Her anime only OG outfit.

No dataset for you as it was heavily padded with the manga and anime screencaps instead it is just the tag summary so people can experiment.

Edit: seems i stripped the metadata of some images I used for normalization, for those ones i think used this as extra just pick one of the ones with the prompt and add it with the respective lora: <lora:SkirtLenght_000000050:-1.8> <lora:microwaist_4_z:.3> <lora:breastsizeslideroffset:0.1> <lora:AtomicHips2:0.1>wide hips,

V1

Anyway The LORA works fine in all the range. I got a couple of weird generations at 512x512 with a top aberration. Might be something I didn't clean properly, but it doesn't seem to appear in higher resolution. I mostly generate in 512x 640,768 or 1024.

Anyway got good results with: <lora:LHMutsumiOtohime:0.7> LHMutsumiOtohime, large_breasts

I tagged the following available outfits:

  • GreenBrownDress (Her default outfit)

  • PurplePartyDress (I think this one is from the ova, pretty nice)

  • OrangeRedYellowDress (random outfit that somehow worked fine, only added it because there was some images of it from several sides making them perfect for training)

  • YellowOrangeDress(Total failure, should be her reddish orange or brown dress with a yellow sweater and maybe a red scarf. only outputs the yellow dress if used without extra tags)

Some useful tags:

  • hair_over_shoulder

  • single_braid (for stability)

  • drill_hair (I pruned this one but the interrogator says that her hair qualifies, should also work to stabilize her hair)

  • <color>_eyes(I left the eye color as selectable so use whatever you want)

  • monochrome(put in the negatives as a lot of doujin pics were used)

I will add the tag summary inside the dataset for people to experiment.

Should be fully NSFW compliant as I used a lot of doujins to pad the dataset.