You should use orph_light to trigger the image generation.
Details
Recommended strength: 1.0 (full). Drop to 0.6–0.8 on faces and on subjects far from the training set.
What it does: paints a scene in dynamic black-and-white value structure, with crisp light and dark masses doing the work, and lands composition most of the time on its own. It is the project's actual product: value-painting turned out to live in the style branch, not the composition one.
Tested with: dpmpp_2s_ancestral / sgm_uniform, 20 steps, guidance 2.5.
