Conventional image generation models do not distinguish between the following two behaviors when given the prompt “looking to the side”:

1. Turning the entire face sideways and rendering a profile view

2. Keeping the face facing forward while only turning the gaze sideways

Additionally, in certain scenes, a strong bias toward rendering the face facing forward can make generating profile views difficult.

This LoRA assists in generating images corresponding to Case 1, where the entire face is turned sideways.

顔を背けているシーンを生成しやすくするLoRA