After reviewing v1, I made minor adjustments as the quality was too low. As shown in the sample, even with just trigger + minimal prompt, it produces reasonably good poses. However, using supplementary prompts like folded, feet, ass, lying, toes, legs up, knees to chest, soles, on back, holding own legs, kneepits, etc., yields more reliable results.

Fortunately, we have great pioneers, but I'm sharing this as one example.

Using the “pretzel pose” prompt can sometimes generate food or mysterious creatures. This LORA was created to prevent that.

The influence of teacher images tends to make the art style realistic. Chairs also tend to appear frequently.

改めてv1を見直すと質が低すぎたので再調整しました。サンプルの通りトリガーと最低限のpromptでもそこそこの打率で出ますが、folded, feet, ass, lying, toes, legs up, knees to chest, soles, on back, holding own legs, kneepits, などで補強した方が確実かと思います。

ありがたいことに偉大な先駆者達がいますが、作例の一つとして公開します。

pretzel poseというpromptを使うと食べ物や謎の生物が生成されることがあります。そうならないために作ったloraです。教師画像の影響で画風はリアルよりになりやすいです。また椅子が出現しがちです。