In standard image generation models, legs folded generally works well most of the time.

However, in certain situations or with larger token sizes, legs may appear spread rather than folded.

This occurs particularly with prompts like pov, missionary.

This LoRA assists in generating images with fully bent legs by training on specific image compositions where legs are completely folded.

Try it when legs don't bend due to biases from other tags.

足を密着するくらい曲げやすくするLoRA。