Image to BW 🐦‍🔥

Most "grayscale" nodes in ComfyUI just average the channels and call it a day. That's how you get muddy, flat mono - dark reds and bright reds collapse into the same gray. This node is different: it's the Photoshop Black & White adjustment layer, ported into your graph, which means it gives you six color sliders instead of one dumb toggle.

The pitch, if you do any print-style work, architectural rendering, or moody B&W photography in ComfyUI: you can darken a blue sky without touching skin tones, or push the cyans so water reads nearly black. That's the whole reason to reach for it.

How it works

The node sorts each pixel's RGB channels and blends them using per-channel weights derived from your six sliders. The math is basically Photoshop's: every pixel contributes a weighted luminance where the weight depends on which hue dominates it, and the "combo" channel (the opposite of the dominant one) gets pulled in too. Long story short: it's a proper hue-aware desaturation, not a weighted average of R+G+B.

The sliders - reds, yellows, greens, cyans, blues, magentas - run from -200 to +300, and note the defaults aren't zero (they're 40/60/40/60/20/80). Positive values brighten everything that hue contributes to; negative values darken it. The unusually wide range is what lets you push past "subtle grade" into "the sky is now black" territory.

Input is a plain IMAGE and output is a plain IMAGE, so you don't need any conversion nodes - wire a loaded image straight in, then out to Preview Image or Save Image. Unlike the OpenCV side of this pack, this node is pure PyTorch and runs wherever your image tensors run.

What to actually do with it

Load an image, wire it in, and drag a slider while previewing. The workflow is boring in a good way: Load Image → Image to BW → Save Image. Where it earns its keep is batch work - if you're generating a grayscale series (say, an architecture portfolio), you can set the six sliders once and get consistent mono across every output instead of whatever the sampler decided gray meant that run.

Install

Same as the rest of the pack: in ComfyUI Manager search ComfyUI-ArchiGraph and install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/vincentfs/ComfyUI-ArchiGraph

Then restart ComfyUI. The pack's install.bat/install.sh pulls in its requirements (OpenCV, matplotlib, numba) even though this particular node needs none of them - the install is heavier than the node. No models to download, ever.

Gotchas

• The sliders are not intuitive "0 = neutral". Defaults are already doing something, so if your mono looks weirdly dark in the blues, that's the 20/60 defaults talking, not a bug.
• You get a single-channel grayscale tensor out. It's a normal IMAGE and previews/saves fine, but if you need RGB later (e.g. to feed a node that insists on three channels), convert before using it downstream.

For architectural and print work where you control the final mono look, this is the B&W node I'd reach for. It's the rare pack utility that does one thing and does it like the desktop app people actually learned on.