PIXEL SPLIT (JOV) 💔

PIXEL SPLIT is the "take it apart" half of Jovimetrix's channel pipeline. One image in, and you get its red, green, blue, and alpha channels out as separate single-channel images, plus a second set of the same channels as masks. Want to blur only the background, boost only the blue, or turn a photo's alpha into something you can edit directly? This is where you start.

It's as close to one job as a node gets. Feed 👾 any image (RGBA, RGB, or even a mask - the pack converts), and you get eight outputs out of it: ❤️ 💚 💙 🤍 as IMAGE (the red, green, blue, and alpha channels rendered as images) and 🟥 🟩 🟦 ⬛ as MASK (the same four channels in mask form). The names are the author's, and they're actually the useful part of the design: the mask outputs are what you plug into masking operations, the image outputs are what you plug into color work or into PIXEL MERGE to reassemble.

Why both? Because in ComfyUI, masks and images behave differently. A mask is single-channel and plugs straight into compositing, inpainting, and other masking nodes; an image is what the color pipeline expects. Jovimetrix gives you both rather than making you convert, and that's a small mercy you'll appreciate the first time you try to wire a channel into a node that insists on one or the other. Also note every output comes out as a list, so a batch of images splits in one pass.

The standard workflow is the round trip: SPLIT an image, do something to one channel (a THRESHOLD on the mask, a TRANSFORM on the image), then feed the modified channels into PIXEL MERGE to get a composite back. A very common beginner win: split, grab the alpha mask, run it through a threshold to clean it up, merge back with a new background color. You can also use a single channel as a monochrome image all on its own - the green channel of a portrait often carries most of the usable luminance detail.

This is a current node, not a legacy one, but its exact output layout has shifted across Jovimetrix versions - at one point the splitter only returned masks, then grew the image outputs, and an old "invert" option got removed along the way. If a downloaded workflow references it with different outputs, check which pack version the workflow was built against. That matters more here than with most nodes.

Install is the whole-pack deal. ComfyUI Manager, search "Jovimetrix", install, restart. Manual route:

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

The dependencies are shared across the pack (numpy, opencv-contrib-python, Pillow, matplotlib, plus the cozy-comfyui helpers) - nothing heavy, no model files. If you're on an older ComfyUI, the pack targets 0.1.3+ and frontend 1.2.40+, so update rather than fight it. And remember the pack's split history: if your workflow's nodes come up missing after an update, GLSL and streaming nodes now live in separate Jovi_* packs; PIXEL SPLIT never left home.