ImageCut

ImageCut does one thing: slices an image across its height and hands you the top and bottom as two separate IMAGE outputs. It's the node for when a tall image needs to be processed in pieces - a character sheet cut in half, a long reference strip split so each half gets its own pass - and then stitched back together.

Two inputs, both floats, both obvious:

• H_cut_ratio (default 0.5) - where the cut lands, as a fraction of the height. 0.5 splits exactly in half; 0.3 puts the cut a third of the way down, top piece being the short one.
• Width_padding (default 0.1) - trims that fraction of the width off each side before cutting. 0.1 shaves 10% off the left and 10% off the right; set it to 0 and you keep full width.

Output is two IMAGEs - upper half and lower half - and you wire each wherever it needs to go.

The source has two edge cases that look like bugs until you know them. If Width_padding is exactly 0, the code explicitly restores full width (so a zero-padding "trim" can't accidentally collapse the slice). And if H_cut_ratio is 0 or 1, both outputs come back as full copies of the input - a degenerate cut that's actually a duplicate. That's the safe behavior (no empty tensors out), just don't be confused when you see it.

Where it fits: cut → process each half differently (different model, different denoise, or one half left untouched) → rejoin with ImageConcat, the sibling node in this pack that stacks two images vertically. Since both halves share the same width, the concat is exact - no rescaling on the way back. That cut-process-rejoin loop is the whole reason this node exists, and it's why the two belong in the same pack.

Install is the pack-standard route. ComfyUI Manager → search "ImgProcessing_ComfyUI" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/nirbhay-faaya/ImgProcessing_ComfyUI
# restart ComfyUI

No models to download, no requirements.txt. The only hard dependency in the pack is opencv-python, and it's imported at the top of the pack's single .py file - if cv2 is missing, ImageCut and all its siblings fail to load together. If the node doesn't appear, that's the first check:

pip install opencv-python   # into your ComfyUI venv, then restart

Like the rest of the pack it's a thin slice of tensor math - no model, no API, nothing to go stale, which matters because the repo is a single commit from May 2024 that the author never touched again. It just keeps working. Feed it a batch and both outputs keep the batch. If you only ever work with single images you can safely ignore every input but H_cut_ratio. Not the most exciting node you'll install this week, but it's exactly the kind of thing you don't want to hand-roll at 2am.