混合式Lora
ハロウィン 魔女衣装で何かしらやってるというコンセプトLora
Trick or Treat つながりで Trick的なヨーヨーとかを適当な気分で追加
メイントリガー式ではないので各状況タグを押さえ、
衣装、人物をそれぞれ変えていく。という使い方となる。
■推奨強度(用途によって強度変化)
お菓子あげ 0.5-
あやとり 0.3-
ヨーヨー 0.8-
薬調合 0.6-
それぞれAdetailer推奨とはなるが、出力品質にそれぞれ幅がある。
解像度は基本的に縦長推奨
元画像意図としては前面上半身となる。
それぞれメイントリガーは無いため、状況要素タグー人物タグや服装タグ部分(サンプル画像参照)を都度変更して使う事。
■特徴その他要素点
お菓子あげ
魔女側からお菓子をくれる構図。おまえがよこすんかーい。
元々は手のひら一杯にお菓子を持った姿を意図してたが割と箱やカボチャが土台になる。適用強度あげれば少なくなるだろうが。
あやとり
元々タグで出るやつ。
ヨーヨー
上から下になげて維持したロングスリーパー(トリック)を意識してるが
構造があまり維持出来ない。どうしてもガチャ気味になっている。
薬調合
混ぜる木のへらをシャモジ指定にしてるが学習の際にタグ追加してなかったため、
表現がやや微妙。
手の持ち手側にすくう側が出てきたり、形状維持に安定感があまり無い。
Mixed-Type LoRA
A Halloween-themed concept LoRA featuring a witch outfit doing various “Trick or Treat”-related actions.
Inspired by the “Trick” part of Trick or Treat, I added some playful actions like yo-yo tricks on a whim.
This is not a main-trigger-type LoRA.
You should combine it with situational tags, and adjust outfit and character tags as needed depending on your intended use.
■ Recommended Strength (Varies by Use Case)
Giving candy: 0.5–
Cat’s cradle: 0.3–
Yo-yo: 0.8–
Potion mixing: 0.6–
Each scenario works best with ADetailer, though output quality varies depending on the case.
Vertical resolutions are generally recommended, as the original images focus on upper-body front views.
Since there’s no main trigger, make sure to modify situational, character, and outfit tags (see sample images) as needed.
■ Features / Additional Notes
Giving candy
A composition where the witch offers candy — “Wait, you’re the one giving it?!”
Originally intended to depict her holding a pile of candy in both hands, but pumpkins or boxes often appear instead.
Increasing LoRA strength reduces this tendency somewhat.
Cat’s cradle
Originally derived from an existing tag-based pose.
Yo-yo
Intended to represent a “long sleeper” yo-yo trick (staying spinning at the bottom),
but structural consistency is weak, and results tend to be somewhat random.
Potion mixing
The stirring tool was meant to be a wooden spatula (shamoji),
but the tag wasn’t properly added during training.
As a result, the depiction is slightly unstable — sometimes the scoop end appears reversed or deformed.
1. The rights to reposted models belong to original creators.
2. Original creators should contact SeaArt.AI staff through official channels to claim their models. We are committed to protecting every creator's rights. Click to Claim
