BD Depth To Shadow Map

A depth map is boring until you realize it's secretly a surface normal map waiting to happen. BD Depth to Shadow Map does exactly that conversion: feed it a depth image and it computes per-pixel lighting - top-down N·L shading plus cavity-based ambient occlusion - and spits out a greyscale "shadow map" that looks like someone actually lit your character. It's a procedural shader input, not a measurement, and for the job it's built for - feeding a skin shader's shadow channel or a PBR derivation pipeline - that's exactly right.

It was designed for the multi-tone skin pipeline in this pack (there's a zombie variant that cranks the AO for sunken eyes and hollow cheeks), but it works anywhere you have depth and want a lighting guide: game-UV prep, stylized shading, "fake a light rig" on a single image.

How the magic happens

Three ingredients get blended per pixel:

1. N·L lighting - it derives a surface normal from the depth gradients and lights it with a direction you control. Default is top-down (light_y = -1), which is why the forehead and brow ridge catch light while sockets fall into shade.
2. Cavity AO - the Laplacian of the depth, which is a fancy way of saying "recesses get dark." Eye sockets, the mouth crease, cheek hollows. ao_radius (8–16 px default) controls what size of hollow counts.
3. Ambient floor (ambient, 0.35) - guarantees nothing goes pure black, so the result stays usable as a multiplier.

light_strength and ao_strength mix those first two (both can go >1 for exaggeration), depth_blur smooths noisy depth before normals are computed, and invert_depth exists because estimators disagree on the "brighter = closer" convention - if your shadows look inverted, flip it.

Inputs: depth (IMAGE, R channel used if RGB; brighter = closer, unless you invert) and an optional mask - pixels outside the mask just get the ambient floor, so a skin mask keeps the shadow map from doing anything weird to the background.

Outputs: shadow_map (the greyscale multiplier - wire it into u_image3.R in the skin shader or any luma-based shadow slot), plus normal_map, cavity_map (handy for debugging what the AO is seeing), and measured_max_curvature (a float for tuning).

Installing

It ships in ComfyUI-BrainDead, so: ComfyUI Manager → search "BrainDead" → install, or

cd ComfyUI/custom_nodes
git clone https://github.com/BizaNator/ComfyUI-BrainDead
cd ComfyUI-BrainDead
pip install -r requirements.txt

then restart. No model downloads - this is pure numpy/torch math on whatever depth you feed it. Upstream, BD Lotus-2 Predict gives you the good diffusion-based depth; DepthAnythingV2 or MiDaS from comfyui_controlnet_aux work fine too if the quality bar is lower.

Where people get burned

The depth convention. If your shadows look "scooped out" backwards, invert_depth is the first toggle, not light_y. Second: this is a guide, not ground truth - the output is meant to be multiplied into a shader channel and then tuned by eye, so don't be surprised that ao_strength at 0.6 looks flat and 1.2 looks like a zombie. Both are intended outcomes.