A LoRA trained on images of people stuck in a wall with an portrait of the said person above it. The only needed trigger is through wall. I do recommend prompting the character features for better results.