The intent of this LoRA when I made it was to cast a shadow onto a wall, from an open door, where beyond the door an activity was taking place that then cast that shadow, indicating the activity.

I thought it didn't work, then I prompted very simply and compact... and it worked.

It does silhouettes easily, no problem, adding detail to the scene will start pulling it to reality but enforcing it with contrasting elements can bring it back into theme, as exampled by my moon-lit scene.

There's probably some interesting, unknown, potential that this LoRA has and I haven't tapped into it yet but I'll keep it active while I produce various hidden, CCTV and security footage scenes to see how well it holds up, and then post my results of course.

I'm looking forward to seeing what other creative minds can do with this, and even in conjunction with other LoRA, which may bring out some magic that we'd have a hard time finding otherwise.

I artificially generated 57 images of silhouettes to create this LoRA and then directed joy-caption to capture the shadowy essence of each. To make these images I just crafted the desired scene and use bright/contrast and desaturate until there was nothing left but black, I used a background remover before that in order to affect only the pixels required, then added a white background back into the mix and fed it into training.