🔱 Artha Color Match

Artha Color Match is the 🔱 Artha pack's heavy artillery for a problem every image pipeline hits eventually: you generated something great, but the colors don't match the reference you had in mind. Film-still energy that comes out greenish, a product shot with the wrong white balance, a character whose palette drifted between passes - this node transfers the color grade from one image to another.

Unlike the simple histogram-matching you'll find in a dozen other packs, this one is a genuine multi-method toolkit. Six color-transfer algorithms, a secondary method you can blend in, gamut control, grain restoration - it's the node you reach for when "close enough" stops being close enough.

How it works

At its core it's the classic color-transfer pipeline: convert both images to a color space, shift the source's statistics toward the target's, and convert back. The depth is in the options. The primary_method dropdown picks the transfer algorithm - Reinhard (statistics-based, the classic), MKL, CIEDE2000, LAB-NN, color-mood, and PDF matching - each with different behavior on tricky grades. iterations (1–10) re-runs the primary method for stronger results. A secondary_method can then be applied and mixed in via one of five mixing_modes (blend, sequential, luminance, detail, or none), with primary_weight controlling how much each contributes.

Three settings prevent the usual color-match disasters: preserve_luminance keeps the source's light/dark structure so you don't get a flat posterized result, gamut_mode sets the working color space (srgb, adobe_rgb, prophoto_rgb, rec2020, dci_p3, or auto), and use_regrain + regrain_strength/regrain_radius restore fine detail after the color shift, which is what keeps the result looking like a photograph instead of a smoothed gradient. strength (0–2) blends the whole result back toward the original.

Inputs that matter

• source_image - the image you're recoloring. target_image - the image whose look you want. This direction trips people up constantly: the target is the reference.
• primary_method - the main algorithm; Reinhard is the safe starting point.
• preserve_luminance - keep it on unless you specifically want the target's lighting too.
• strength - your safety dial. Start at 1.0, back it off to ~0.6 if the grade looks overcooked.

Output is a single image IMAGE output, ready to feed a save/preview node or back into an img2img pass.

Installing it

One-pack install, same as the rest of Artha:

cd ComfyUI/custom_nodes
git clone https://github.com/Cyrostar/ComfyUI-Artha-Nodes
pip install -r ComfyUI/custom_nodes/ComfyUI-Artha-Nodes/requirements.txt

then restart, or install "ComfyUI-Artha-Nodes" through ComfyUI Manager. The color matching leans on scikit-learn and opencv-python from the pack's requirements - those are the real dependencies here, and they're heavier than the pack's other nodes. No API key, no model downloads, fully local.

Common issues

• Result looks washed out or flat: usually preserve_luminance off, or an aggressive iterations setting. Re-enable luminance and drop back to 1–2 iterations.
• Green/gray cast after matching: a gamut mismatch. Try switching gamut_mode to match the target's working space, or auto.
• Grainy, splotchy detail: turn on use_regrain and tune regrain_radius (3 is the default) to pull source detail back.
• Tiny color shift, not enough: crank strength toward 2.0, or add a second method via secondary_method.

The one-liner: it's the pack's most technical node and it repays the fiddling. If all you need is "make image B look like image A, quick", the Reinhard default with luminance preserved gets you 80% there - the extra knobs are for when 80% isn't good enough.