Samples weren't promising
Currently this seems to recognize the concept when a tree is involved. Characters often face the "wrong way", with their backs against the object in the output.
Probably needs more images in training if I ever update this. This was trained on 10 images only


