Change Image Border

The name tells a small lie you only notice after the run: Change Image Border doesn't pad your image with a strip of color. It recolors the outer border_size pixels that are already there, and hands you back an image the exact same size. Dimensions in = dimensions out, just with a frame painted over the edge. It's a cosmetic utility, and honestly that's all it's trying to be. That's not a criticism - sometimes you just want a thin black frame on a batch of outputs, and a dedicated node beats firing up an editor.

What it's for

Reach for it when you want a uniform edge on finished images:

• Framing batch generations so a grid or contact sheet doesn't blur into itself
• Giving preview thumbnails a consistent dark edge before you drop them into a gallery, a video, or a composite
• Quick "vignette-ish" darkening of an image's edge without leaving ComfyUI

But if what you actually want is to grow the image with padding - room for outpainting, or a caption strip along the bottom - this is the wrong node. Core ComfyUI's Pad Image nodes do that by resizing the canvas. ChangeImageBorder is a picture frame, not a mat, and confusing the two is the easiest way to end up confused.

How it works

The whole pack is one ~100-line file, and the mechanism is plain pixel poking. Each input image becomes a PIL image, gets converted to RGBA, and numpy slices the top, bottom, left, and right strips of border_size pixels. Those strips get overwritten with [value, value, value, 255] - your color at full alpha - and the result is converted back to a tensor. Batches work: feed it ten images and it loops through each and concatenates them back together. No models, no weights, no downloads, no dependencies beyond numpy/PIL/torch that ComfyUI already ships. This is as close to dependency-free as a custom node gets, which in the ecosystem's usual dependency hell is genuinely refreshing.

The inputs that matter

• image - any IMAGE tensor you have around.
• value (INT, default 0) - the border color. Here's the catch: it's a single grayscale value from 0–255, applied identically to the red, green, and blue channels. 0 is black, 255 is white, everything between is some gray. There is no hue input, so you cannot make a red or blue border with this node. If you need a colored frame, you'll be using ImageCompositeMasked or a color node instead.
• border_size (INT, default 2) - the thickness of the frame in pixels. A 2 reads as a thin outline; crank it to 10+ on a big image and it starts to look like a real frame.

Output is image, same IMAGE type. The one subtle gotcha: the node converts internally to RGBA, so your 3-channel RGB input comes back as 4-channel RGBA with a solid alpha. Most downstream nodes shrug at that, but a few expect exactly three channels and will throw an odd error - worth knowing before you wonder why your pipeline suddenly grew a channel.

Install

The standard two ways:

1. ComfyUI Manager → Install Custom Nodes → search for Change Image Border (pack name ComfyUI_Change_IMAGE_BOREDER) → install and restart.
2. Manual:

   cd ComfyUI/custom_nodes
   git clone https://github.com/githubYiheng/ComfyUI_Change_IMAGE_BOREDER
   

   Then restart ComfyUI. No requirements.txt, no model files - the pack is a single Python file, so there's nothing to go wrong on install.

Gotchas & troubleshooting

• It prints aaaaa and bbbb to your console every run. Harmless debug leftovers from the author. Cute, slightly embarrassing, and a good reminder this is a quick utility rather than a polished library. (The pack's README has since been deleted - git history says "fix: remove readme" - so the code is the only documentation, and it's accurate.)
• Border color is gray-only. If you expected a full color picker, the single value INT is all you get.
• Overwrite, not pad. If your image didn't grow, that's correct behavior. If you wanted it to grow, you need an image-pad node.
• Node missing after install? Restart ComfyUI, and confirm the folder landed in custom_nodes/ with the __init__.py intact.

It's a small tool with one honest job. If that job matches your workflow - a clean frame on a batch of outputs - it'll do it without fuss and without dragging in a single extra dependency.