BD Color Extract

Sometimes you don't need a segmentation model to find the thing - the thing is just a color. A blue viseme atlas where you need the blue pixels. A painted shadow region that's a specific hue. A wardrobe item that's a flat color on a transparent render. BD ColorExtract finds every pixel close to a target color and hands you a mask. Zero model load, zero inference, milliseconds - and it's the kind of deterministic op the KB's post-processing doc would tell you to reach for instead of another diffusion pass.

The default is set up for the pack's own use case: target_b is 1.0 and the mode is hue, so out of the box it extracts blue. The defaults literally spell out "Default 0.0 (for blue)" in the tooltips. Change the RGB floats to pick any color.

The inputs that matter

• image - the source.
• target_r / target_g / target_b - the target color, each 0–1. Defaults are blue (0, 0, 1).
• match_mode - the heart of it. hue (default) matches by color identity in HSV, ignoring brightness - "any blue, dark or light." That's the one for painted shadow extraction, where the same hue appears across many brightness levels. rgb_distance matches a strict RGB triplet within a tolerance - for when you want only pixels very near one exact color.
• tolerance - how loose the match is. In hue mode, 0.15 ≈ ±54° of hue. Lower = stricter.
• min_saturation - hue mode only: pixels below this saturation are ignored, so greys and near-whites don't falsely match the target hue (their hue is noise). Set 0 to include them.
• soft_edge - smoothstep falloff around the tolerance boundary, 0 = hard binary cut.
• gradient_mode - ON gives a continuous mask scaled by how close each pixel is to the target color (target = 1.0, far = 0.0). The "how blue is this pixel" extraction, useful for tint maps. OFF gives the thresholded mask.
• invert - flip to extract everything except the color.
• mask (optional) - restrict extraction to a region; output is zero outside it, so you can limit color matching to skin areas, for example.

Outputs come in both flavors: color_mask (a 3-channel greyscale image, R=G=B=mask value) and mask (a single-channel MASK for nodes that want that type). You don't have to convert - both are there.

The workflow shape

[render/atlas] → [BD ColorExtract] → mask → [inpaint / composite / mask-flatten / pack]
                 target: your color
                 match_mode: hue

A natural pairing is with the pack's other segmentation outputs: use a mask from this node to drive a crop, an inpaint region, or feed a mask-flatten for a game texture. Because it's a lookup-style op it runs instantly, so you can afford to tune tolerance and soft_edge live and watch the mask update.

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

Hue mode is powerful and has a trap: pure greys and whites have unreliable hue, which is exactly why min_saturation defaults to 0.1 - if you're getting noise from neutral areas, raise it. And remember hue-mode matching is brightness-blind by design, so if you actually need "that exact mid-blue and not the dark navy version," switch to rgb_distance and tighten tolerance. Two output channels can confuse people at first - a node wanting IMAGE gets color_mask, a node wanting MASK gets mask; they're the same data in two wrappers. Finally, on very desaturated or color-graded sources, tune tolerance from a known pixel - guessing the value blind is how you get a mask that covers half the image.