Conventional image generation models exhibit a strong bias toward the keyword “punch,” resulting in fists being generated prominently toward the viewer.

Due to this bias, while post-punch expressions are easily achievable, depicting the pose of a raised fist just before striking is challenging.

This LoRA assists in generating poses of a raised fist, capturing the moment immediately before an attack.

パンチ直前の拳を振りかぶったシーンを生成するLoRA