This LoKR can be used at strengths 1 - 1.75. You should prompt to indicate where Squidward is looking as the training dataset is very small and lacks viewpoint diversity. However, the model itself is very flexible and can do illustrated, stylized, and realistic scenes. Works in Comfy but may not work in other generators that do not support LoKRs.

Trained using perceptual depth anchoring with ai-toolkit-perceptual.