BD Center Median Luma

Some shaders are picky about where your image's brightness sits. The V-curve skin shaders this pack's GLSL pipeline targets assume the input's median luma lands at the curve's midpoint - the u_float0 uniform - so the tint dominates evenly across shadows and highlights. Real generated images don't cooperate; a face that reads too dark overall pushes the median off-center and the whole tint lands wrong. BD Center Median Luma fixes the position: it computes the luma median of your input, then adds a single constant offset so the median lands exactly on your target_center. The node description is precise about what it does and doesn't do: it preserves the histogram shape (spread and contrast unchanged) and only shifts position. No contrast remap, no tone curve - a straight additive shift.

This is deliberately the other half of a two-node pair. BD Normalize Luma controls the spread (range); this node controls the position (median). The description says it plainly: normalize first to set the spread, then center to set the median position, then feed the result to the shader.

The inputs that matter

• image - the source, single or batch. Each frame is centered independently.
• mask - strongly recommended. The median should be measured from the subject, not the background. Without a mask, background pixels bias the measurement. Wire a skin or body mask here and the defaults handle the rest.
• target_center - where the median should land, default 0.5. Set it to your V-curve midpoint (u_float0).
• luma_standard - the RGB→luma weights: bt709 (default, modern sRGB/Rec.709, right for Unity/Unreal), bt601 (legacy NTSC, matches Photoshop Desaturate), or average. Pick bt709 unless you have a specific reason not to.
• statistic - median (default) is robust to cheek highlights and deep shadows; mean is smoother but outlier-sensitive. Median is the right call for V-curve centering.
• calc_from_mask - which pixels are used to measure. ON (default) measures mask pixels only. OFF measures the whole image, with exclude_transparent as a fallback for images that already carry a body-shaped alpha.
• apply_to_mask_only - where the shift is applied. ON (default) shifts only within the mask, leaving the background untouched. OFF shifts everything.

Those last two are independent, and the combos are worth a moment: you can measure from the mask but apply to the whole image, or measure the whole image and apply only in the mask. The defaults (measure and apply in-mask) are right for most cases.

Outputs: image (the centered result), measured_median (FLOAT, what the input's median actually was), and shift_applied (FLOAT, the offset added - negative if the input was too bright, positive if too dark). Both floats are great for wiring into diagnostics or feeding the shader's uniform directly.

Installing it

Part of BizaNator/ComfyUI-BrainDead (BrainDeadGuild). 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

Restart; it's under 🧠BrainDead/Segmentation.

Gotchas

The classic failure is skipping the mask. Without it, a dark or bright background drags the measured median off, and the "centered" result is still off-center - because the shift was calibrated on the wrong population. If your source already has a proper alpha, calc_from_mask=OFF + exclude_transparent=True is the honest fallback. Second: this node shifts position only. If your image is also too flat or too contrasty, center it and then normalize the spread - or the other way around; the order matters less than doing both. And one subtlety: if your shader expects a mean rather than a median, flip the statistic - the two disagree exactly when your image has outlier highlights, which skin often does.