REMADE Batch Image Select Channel

The debugging-adjacent utility of the pack. REMADE Batch Image Select Channel takes a batch of images, splits each into its RGB channels, keeps the one you choose (red, green, or blue), and returns the batch with that channel's data as a monochrome image - same channel tripled back into RGB. It's the "show me what the red channel actually looks like" node, scaled to a whole batch.

When is that worth a node? Channel isolation is a genuinely useful diagnostic and preprocessing trick. Look at the individual channels of a noisy or compressed image and you'll often spot where the artifacts live (blue channels carry the worst noise in most phone photos). A single channel can also make a decent mask in a pinch - the red channel of a skin-tone image is a rough luminance-ish map, for instance. And if you're building a control-signal pipeline, isolating a channel gives you a grayscale-ish image you can threshold into something usable. Mostly, though, this is a "peek under the hood" node, and it's honest about being that.

How it works

For each image in the batch: convert to RGB, split() to get the three channels, grab the chosen one, convert it to grayscale, then merge it back into an RGB image with the same channel in all three slots. The output looks grayscale - a red-dominant image read through this node's red channel will be bright where red is strong and dark elsewhere, so it reads like a luminance map of that channel. Technically the output is still RGB (each R=G=B), not a true single-channel image or a MASK.

Inputs and outputs

• images (IMAGE) - the batch.
• channel (enum: red, green, blue) - which channel to keep.
• Output: IMAGE - same batch length, each frame the chosen channel tripled into RGB.

The catch

It returns RGB, not a mask and not a raw channel. If you were hoping to wire this into a MASK input, it won't connect - you'll need a Convert-Image-to-Mask step (core ComfyUI has one) in between. And because the output is a tripled RGB image rather than a single-channel tensor, the actual channel data is still there, just dressed up as a gray image. That's fine for viewing and for most preprocessing; it's just worth knowing the output isn't a true one-channel tensor.

Installing it

One of 12 nodes in 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; numpy declared (with the pack-wide OpenCV caveat - pip install opencv-python if the pack won't import, thanks to the color-blend node).

It's a one-knob utility with a narrow job, and that's okay - every toolbox has a channel-viewer. Just remember the output is RGB pretending to be grayscale, and convert to a mask if that's what you actually need.