七夢
katana held vertically beside head two handed hasso guard pose,
LoRA名：（説明必読・特殊条件）DiT2 八双の構え pose V2.0
１・はじめに
・このポーズLoRAは、以下の２点に関するポーズLoRAとなります。 ①刀を装備する。 ②八双の構えをとる。
・トリガーの状態で生成をかけると、女の子が刀をもって八双の構えをとっている画像が生成されます。
全身を表示させる場合には、「Take a full-body image that shows all the way down to your feet.」等の全身を示すプロンプトを追加でご入力ください。
２・使い方
・基本的な使用例としては、キャラクタープロンプトの中に、トリガーワードを自然な感じで入れる感じです。
プロンプト例題 「A girl with asymmetrical bob hair in brown with bright yellow highlights, a hair accessory, and orange eyes.katana held vertically beside head two handed hasso guard pose,Take a full-body image that shows all the way down to your feet.」
例題訳 「茶色の非対称ボブヘアに明るい黄色のハイライト、ヘアアクセサリー、オレンジ色の目をした女の子。刀を頭の横に垂直に持ち、両手で八相構えのポーズをとっています。足元まで写る全身写真を撮ってください。」
⭐重要な点⭐ 上記だけでは、生成成功率は「一括×４」生成で、４枚中０～２枚 という感じになってしまいます。必ず、以下の２つの操作を行ってください。GPT式、徹底的に頼み込む作戦です。これで基本的には安定します。※追加プロンプトは状況に応じて減らすのもアリだとは思いますが、減らした状態でのテストはしていません。
🍓プロンプトに以下を追加する。 「hasso no kamae,traditional Japanese katana stance,katana held vertically beside the head,blade rising straight upward beside the face,two handed katana grip,both hands gripping the katana handle,hilt near the right shoulder,raised elbows,upper body slightly twisted,battle ready stance,feet firmly planted,full body,front view,standing pose,single character,katana clearly visible,hands clearly visible, arms clearly visible,serious expression,focused eyes,samurai combat posture,」
🍈ネガティブプロンプトを１度削除し、以下を貼り付ける。 「one handed sword,holding sword down,sword pointed down, sword at waist,sword resting on shoulder,sword behind back, sword in front of body,low guard,middle guard,overhead guard,relaxed standing pose,open arms,wrong hand position,extra sword,multiple swords,missing hands,bad hands,broken fingers,hidden hands,cropped sword,」
３・他のLoRAとの併用の方法
・他のLoRAと本LoRAを選択して頂き、同じ強さ（０．７）の生成で も問題なくポーズを取らせるケースも存在しますが、併用するLoRA次第です。
本ポーズLoRAは、他のキャラLoRA・スタイルLoRA等とあわせると その影響を強く受けますので、ポーズを保てないことが多いです。
そのため、重要な点に記した事項に関する対応を行った上での生成 が必要となります。
その他、同じ強さ（０．７）で上手くいかない場合には、LoRAの強さに関する調整でご対応下さい。基本的には併用LoRAが増えるほど難度があがるように思います。
🍉強さ設定の例 〇本LoRA１．０、キャラLoRA０．７、スタイルLoRA０．７ 上記で難しいようでしたら、キャラLoRA、スタイルLoRAの強さを 下げる感じです。ここは本当に併用するLoRA次第で最適な設定は変化します。一気に０．２まで下げてしまうと、明確に変化するケースが多いような気がしています。
４・その他
トリガーワードの「katana」を「Rifle」に置き換えて頂くと、刀でなく銃を装備した状態で、八双の構えをとってくれるケースも存在します。
５・ご参考（LoRAの作成について）
・ポーズLoRAが初めてだったこともあり、本LoRAの作成は１度失敗をしています。当初、様々な方向から見た「八双の構え」の資料を作成した上でそれらを読み込ませ、LoRAを作成しましたが、「様々な方向」がダメだったようで、「八双の構え」を基本的には取らないLoRAとなってしまいました。そのため本LoRA作成の際には、すべて正面から見た、「八双の構え」の女の子で統一して、LoRAを作成したところ上手くいきました。使用枚数は約３０枚×３セット＝約９０枚。で作っています。
扱いずらいLoRAですが、少しでも楽しんで頂けたのであれば嬉しいですm(__)m
(Description Must Read - Special Conditions) Sword Stance Pose LoRA
LoRA Name: (Description Must Read - Special Conditions) DiT2 Hasso Stance Pose V2.0
This pose LoRA relates to the following two points:
① Equipping a sword.
② Assumeing the Hasso stance.
When generated with the trigger in place, an image of a girl holding a sword and assuming the Hasso stance will be generated.
To display the full body, please add a prompt indicating the full body, such as "Take a full-body image that shows all the way down to your feet."
A basic usage example is to naturally insert the trigger word within the character prompt.
Prompt Example: "A girl with asymmetrical bob hair in brown with bright yellow highlights, a hair accessory, and orange eyes. She holds a katana vertically beside her head and assumes a two-handed hasso guard pose. Take a full-body image that shows all the way down to your feet."
Example Translation: "A girl with asymmetrical brown bob hair in brown with bright yellow highlights, a hair accessory, and orange eyes. She holds a katana vertically beside her head and assumes a hasso guard pose with both hands. Please take a full-body image that shows all the way down to your feet."
⭐Important Point⭐ With only the above, the generation success rate for "Batch x 4" generation will be around 0-2 out of 4 images. You must perform the following two operations. This is the GPT method, a strategy of persistent pleading. This will basically ensure stability. *While it may be possible to reduce the number of additional prompts depending on the situation, we have not tested it with fewer prompts.
🍓Add the following to the prompt.
"Hasso no kamae, traditional Japanese katana stance, katana held vertically beside the head, blade rising straight upward beside the face, two-handed katana grip, both hands gripping the katana handle, hilt near the right shoulder, raised elbows, upper body slightly twisted, battle ready stance, feet firmly planted, full body, front view, standing pose, single character, katana clearly visible, hands clearly visible, arms clearly visible, serious expression, focused eyes, samurai combat posture,"
🍈 Delete the negative prompt once, then paste the following.
"One-handed sword, holding sword down, sword pointed down, sword at waist, sword resting on shoulder, sword behind back, sword in front of body, low guard, middle guard, overhead guard, relaxed standing pose, open arms, wrong hand position, extra sword, multiple swords, missing hands, bad hands, broken fingers, hidden hands, cropped sword,"
While it's possible to select this LoRA and other LoRAs and generate them at the same strength (0.7) without problems, it depends on the LoRAs being used.
This pose LoRA is strongly affected when combined with other character LoRAs, style LoRAs, etc., and often fails to maintain its pose.
Therefore, generation requires addressing the points mentioned in the important section.
Furthermore, if it doesn't work with the same strength (0.7), please adjust the strength of the LoRA. Basically, the more LoRAs you use together, the more difficult it becomes.
🍉Example of Strength Settings 〇Main LoRA 1.0, Character LoRA 0.7, Style LoRA 0.7 If the above is too difficult, try lowering the strength of the Character LoRA and Style LoRA. The optimal setting really depends on the LoRAs you use together. I feel that lowering it all the way down to 0.2 often results in a noticeable change.
If you replace the trigger word "katana" with "Rifle," there are cases where the character will assume the Hassō stance even when equipped with a gun instead of a katana.
・Since this was my first time creating a pose LoRA, I failed the first time I created the main LoRA. Initially, I created reference materials of the "Hassō stance" from various angles and imported them to create the LoRA, but it seems that the "various angles" were the problem, resulting in a LoRA that basically did not assume the "Hassō stance." Therefore, when creating this LoRA, I decided to use only images of girls in the "Hasso no Kamae" stance, viewed from the front, and it worked well. I used approximately 30 images x 3 sets = approximately 90 images.
Although LoRA is difficult to work with, I would be happy if you enjoyed it even a little m(__)m
![Henshin Pose [Pony] - AI Model cover image](https://images-ng.pixai.art/images/thumb/24b98600-376e-4039-9984-76ce70ce961b)
