When you prompt a standard image generation model with looking down, it may exhibit a strong bias toward from below-related outputs.

Furthermore, since the looking down tag is trained without distinguishing between “the face is looking down” and “only the gaze is looking down,” it can be difficult to consistently produce the pose the user desires.

This LoRA supports generating images where the face is looking downwards without the from below bias.

顔が下を向いている画像を生成しやすくするLoRA