LayerUtility: ImageChannelSplit

Splits one image into its individual color channels, each returned as its own grayscale-looking image. It's a Photoshop-era move that most ComfyUI workflows never need - until you specifically want to isolate the luminance from the color, fix a color cast on just one channel, or feed a single channel (say, the L in LAB, which is pure brightness with no color information) into something that wants a grayscale mask rather than a full-color image.

Most diffusion workflows stay in full-color RGB the entire time and never touch this, so if you've never needed a node like this you're not missing anything - it's specifically for the compositing-heavy corner of the pack, where you're treating an image more like a Photoshop document than a generation output.

How it works

You give it an image and a mode - which color space to split it into - and it decomposes the image into that space's component channels, each returned as a separate single-channel image. RGBA gives you red, green, blue, and alpha. YCbCr gives you luminance (Y) and two chroma channels - the classic trick for isolating brightness from color. LAB gives you lightness (L) plus the a/b color-opponent channels. HSV gives you hue, saturation, and value as separate images.

The inputs and outputs that matter

• image - what you're splitting.
• mode - the color space: RGBA, YCbCr, LAB, or HSV. Pick based on what you actually need isolated - YCbCr or LAB if you want brightness cleanly separated from color, RGBA for straightforward per-channel RGB work, HSV if you specifically want hue or saturation on its own.

Outputs: channel_1 through channel_4, each an IMAGE. For RGBA that's R, G, B, A in order; for the three-channel modes (YCbCr, LAB, HSV), channel_4 is unused/empty - only the first three carry real data.

How to install it

Search ComfyUI Layer Style in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/chflame163/ComfyUI_LayerStyle
pip install -r ComfyUI_LayerStyle/requirements.txt

Restart; it's under 😺dzNodes → LayerUtility.

Common issues

The main point of confusion is picking the wrong mode for the job - if you want "just the brightness, no color," RGBA's individual channels won't give you that (each RGB channel still carries some brightness and color information); you want YCbCr's Y channel or LAB's L channel instead. People also sometimes expect channel_4 to be populated for three-channel modes and are confused when it's blank - that's expected, not a bug, since YCbCr/LAB/HSV simply don't have a fourth channel to report. And remember this node pairs with ImageChannelMerge in the same pack - if your plan is split, edit one channel, recombine, keep the mode identical on both ends or the reassembled image will come out visibly wrong. If you've split into LAB specifically to boost contrast or fix exposure on just the L channel, do that edit carefully - LAB's a/b channels are unintuitive to look at directly (they encode color-opponent values, not anything that reads visually as "color" on their own), so don't be alarmed if channel_2/channel_3 look like flat gray-green noise; that's normal for LAB, not a broken split.