majogari mix

・NAIを含まないイラストモデル。
・主にtrainDifference機能で、学習モデルと学習モデル同士のマージモデルのみを使用。
・個性的な色彩、人気モデルとは異なったキャラクターデザインなど。
・混ぜたモデルの傾向から、おそらくはファンタジーも得意なはずだが、よく分からない。
・前髪に偏りがあるので、そこを変更するプロンプトを知っていれば役に立つかも。
（swept bangs、blunt bangsなど）
・Danbooruタグは利用可能だが、やや癖があり。弱いと感じたら強度を上げると効くはず。
・プロンプトによっては、背景が強すぎる傾向があり、背景の強度を下げると人物の比重が増す。
・????も出せるが、どの程度優れているかは分からない。
・作例は少女、城、水彩少女、ドラゴン。

・Illustration model without NAI.
・Only training models and merged models between training models were used, mainly in the trainDifference function.
・Unique colors, character designs that differ from popular models, etc.
・The tendency of the mixed models suggests that perhaps fantasy is also a strong suit, but I'm not sure.
・Because the bangs in the image are biased, it might be useful if you knew the prompt to change it.
(swept bangs, blunt bangs, etc.)
・Danbooru tags can be used, but there are some peculiarities. If you feel weak, increase the intensity and it should work.
・Some prompts tend to have too strong a background, and reducing the background strength increases the weight of the person.
・Examples are a girl, a castle, a watercolor girl, and a dragon.
・This can also generate ????, but I don't know how good it is.
・English depends on translation tools.

・ヒント、あるいは個人的な用法／Tips or personal use

Sampler:Euler a
CFG scale:7.5（5～12？）
Sampling steps:20～
Clip Skip:2
full bodyはHires.fixを使用。
VAEは使用者の自由に。
Use Hires.fix for full body etc.
VAE is your choice.
Negative:law quality, (worst quality:1.2), lowres
これで十分ですが、bad_handsなどを追記すればもっと良くなります。
必要なら、EasyNegativeのようなembeddingsを用いてもいいでしょう。
This is good enough, but adding bad_hands etc. would make it even better.
You can use embeddings like EasyNegative if you want.

・サンプルのプロンプト／sample prompt

少女／girl
1girl, blue hair, semi-short hair, serious (girly) look, younger sister, standing close-up, [Fashion like a magazine]
Negative:law quality, (worst quality:1.2), extra_arms, text, disembodied, fused fingers, lowres, cropped, bad_hands, poorly drawn hands
城／castle
8k wallpaper, (photo realistic), highres, landscape, beautiful castle, depth of field, sharp focus
Negative:law quality, (worst quality:1.2), Simple Background, text, logo, human, back view, face, [3D]
水彩少女／watercolor girl
(watercolor ambient:1.2), (Medieval town), 8k wallpaper, 1girl, black Witch Student, innocent smile, Pale pink_hair, green witch hat, (ultra-detailed), sharp focus
Negative:law quality, (worst quality:1.2), extra_arms, text, disembodied, fused fingers, lowres, cropped, bad_hands, poorly drawn hands
ドラゴン／dragon
8k wallpaper, (TCG art), highres, [fantasy] (dragon:1.2), wingd dinosaur, three color
Negative:law quality, (worst quality:1.2), Simple Background, text, logo, human, back view, face, [3D]