Our favorite captain from the Nadesico. She is the last of my planned Nadesico characters as Inez Fressange and Erina Kinjo Won are already complete. Interestingly Yurika is like a yandere but with zero violent tendencies. Instead of killing your harem she would help you recruit and manage it :P The lora is trained mostly on Watanabe Yoshimasa's and Hell And Heaven's work. So support those artists if you can. Specially Watanabe Yoshimasa who has made dozens of great Nadesico doujins.

V2

Erm... why did i retrain this? Honestly I as confused a you are. V1 works fine and i find it more artistic. I mostly wanted to train the LOCON version but ended up tweaking the tagging so I just trained both with the changed dataset.

Look at the attached XYZ plot so you get an idea of the weights results.

Got nice results with <lora:YurikaMisumaruV2:0.75> yurikamisumaru,

.75 weight seems to be some sort of threshold as at .7 it looks more generic and at .8 much more like the anime doujins. This should work ok at 1 but it will display the style heavily.

As i said in V1, images of her out of uniform are scarce so i didn't bother to add triggers for them only 2 are properly added:

• Military_uniform_White_turtleneck_jacket_Black_pencil_skirt_Black_pantyhose_White_capelet: Her captain outfit.

• CaptainHat: her hat.

As Always the tag summary is inside the dataset so get it and experiment.

V1

The Lora is rather forgiving and works fine with: <lora:YurikaMisumaru:0.75>, YurikaMisumaru, CaptainUniform,

It seems stable from .3 to 1.2, at 1.3 it is very interesting and somewhat artistic(not good for general use but nice in it's own way).

As for her outfits, images of her out of uniform are scarce so i didn't bother to add triggers for them only 2 are properly added:

• CaptainUniform

• CaptainHat

You may want to set her hair color and eye color as it varies and a good part of the dataset is in monochrome. Honestly I am confused if her eyes are green or blue and her hair changes from blue to purple whenever, so YMMV just pick what you like.

You might be able to possible make the lora spit her idol costume, wedding_dress and naked_apron with some coaxing as they are inside the dataset.

All in all a very nice and mild tempered LORA, I seem to have a harder time when using screencaps of 90s anime than with modern stuff or artwork. Might be all the noise and graininess they have even after filtering them. The ones I have made mostly from screencaps have been a bit of a nightmare(Erina Kinjo Won and Ifurita... I also blame their lipstick).

I'll add the tag summary with the dataset so you can experiment.

Some interesting tags from the dataset:

• white_gloves

• military_uniform

• thighhighs

• pantyhose

• garter_belt

• garter_straps

• capelet

• stockings

• pencil_skirt

• school_uniform

• wedding_dress