The Japanese explanation is followed by a machine-translated English explanation.
https://pixai.art/ja/model/1641598710990432265
↑上記のURL先のSD1.5モデル「funifunisitai_inVAE」の絵柄再現を試みたLoRAです。比率は0.7で効きが悪ければ1.0以上にまで引き上げても大丈夫です。モデルごとによってかなり効きが変わります。また、版権キャラタグを使うと絵柄負けを起こしがちです。
白髪ショートヘアの頭部のみを学習画像に入れているので色々な髪型や目の色指定に反応します。ただし表情の感情指定に関しては喜びと悲しみに学習画像を寄せたため、比率を上げ過ぎると無表情や微妙な感情表現に反応しない場合があります。
Tsubaki.2版のトリガーワード「rounded proportions」を入れると体型全体がぷにぷにします。ぽっちゃり体型にしたくない人は外してください。
https://pixai.art/ja/model/1641598710990432265
This LoRA attempts to reproduce the art style of the SD1.5 model "funifunisitai_inVAE" at the URL above. The ratio is 0.7, but you can increase it to 1.0 or higher if it's not effective enough. The effectiveness varies considerably depending on the model. Also, using copyrighted character tags tends to result in the art style being lost.
Only a head with short white hair is used as the training image, so it responds to various hairstyles and eye color specifications. However, regarding facial expression emotion specifications, the training images are geared towards joy and sadness, so increasing the ratio too much may result in it not responding to expressionless faces or subtle emotional expressions.
Entering the trigger word "rounded proportions" in Tsubaki.2 version will make the entire body shape squishy. If you don't want a chubby body shape, remove it.
