REMADE Batch Color Blend

The most interesting node in the pack, and the only one that brings a real algorithm with it. REMADE Batch Color Blend does color transfer between pairs of images using the classic Photoshop blend-mode family: Hue, Saturation, Color, and Luminosity. Give it a batch of blend_images and a batch of base_images, pick a mode, and it merges one channel from the blend image into the base image - across the whole batch in one pass.

What that buys you in practice: color grading, plain and simple. Take the hue/saturation from one image (a reference, a look, a mood board color) and stamp it onto another image's shape. The classic use is "keep my generated image's detail, but make it match this reference's palette" - the Color mode is the workhorse there, replacing hue and saturation while keeping the base image's luminance (so the shadows and highlights stay put).

How it works

The guts are a hand-rolled RGB↔HSY conversion - not OpenCV's HSL, and not Photoshop's exact math, but a proper hue/saturation/luma model with the standard YIQ luma weights (R=0.299, G=0.587, B=0.114). For each pair it converts both images to HSY, then per mode takes one component from the blend image and the other two from the base:

• Hue - blend's hue, base's saturation + luma. Recolors without changing the base's richness or brightness.
• Saturation - blend's saturation, base's hue + luma. Richness swap, no color change.
• Color - blend's hue + saturation, base's luma. The full "recolor while keeping lighting" mode.
• Luminosity - blend's luma, base's hue + saturation. Shape/lighting transfer, no color change.

The whole thing runs on OpenCV (cv2), which is where the installation trap lives.

Inputs and outputs

• blend_images (IMAGE) - the batch whose color/luma gets borrowed.
• base_images (IMAGE) - the batch that gets modified. Batch sizes must match (the node asserts it).
• mode (enum: Hue / Saturation / Color / Luminosity) - pick one.
• Output: IMAGE - one blended frame per pair.

The gotchas

• cv2 is required but undeclared. requirements.txt lists only numpy, yet this module does import cv2 at the top of the file - and since __init__.py imports every module on pack load, a missing OpenCV takes down the entire pack, not just this node. If Remade_nodes fails to import, pip install opencv-python in your ComfyUI environment is the fix. Most installs already have it (something else usually pulled it in), but it's the first thing to check.
• Custom HSY ≠ Photoshop. The hue math here is decent but its own flavor - don't expect pixel-identical results to Photoshop's blend modes on the same pair of images. Close enough for grading, not for matching a reference file exactly.
• Batch sizes must match, and both images are expected in the same dimensions (the node doesn't resize). Mismatched batches throw an assert error mid-run.

Installing it

Part of Remade_nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Pheat-AI/Remade_nodes

Restart ComfyUI, or search Remade_nodes in ComfyUI Manager. No models. Just the OpenCV dependency to be aware of.

Of all the pack's nodes, this is the one that could plausibly replace a stock image-editing step in a production workflow. Batch color transfer with four sensible modes and no VRAM cost - make sure cv2 is installed, then let Color mode do the heavy lifting.