M Int Sub (a - b)

Addition gets the attention, but half the useful arithmetic in a ComfyUI graph is subtraction: stepping a seed back, shrinking a dimension, computing a difference. M Int Sub is the mirror image of the pack's add node - a - b, one integer out, nothing clever going on.

Two INT inputs, a and b (default 0, signed 32-bit range), and one output named a_minus_b. The implementation is literally Python's a - b. Boring in the best way - there's no mode switch, no rounding surprise, no hidden behavior to learn before you can trust it.

The genuinely useful patterns:

• Backwards seed stepping. a_minus_b wired into a converted KSampler seed input gives you deterministic "previous seed" style variation - the exact inverse of the "+1 per run" trick.
• Residual math. In upscaling or latent workflows, "target minus current" (target size minus current width, budget minus steps used) is a recurring shape, and having it as a visible node keeps the intent in the graph instead of buried in a value you hand-computed.
• Offsets that need to stay coupled. When two values must keep a fixed difference and you retune one, feeding both from a single subtraction keeps them locked together.

Because the sockets accept wires or typed values and compose with the pack's other math nodes, M Int Sub is one of those nodes you'll drag in without thinking once you've got the pack installed. It handles negatives correctly (subtracting a negative adds), so the "reverse the sign" trick you'd expect works.

It ships in ComfyUiNodes (repo Madlumi/ComfyUiNodes, author "Madanie"), a lightweight grab-bag of integer math, string helpers, and workflow-glue nodes. Install via ComfyUI Manager - search "ComfyUiNodes" - or:

cd ComfyUI/custom_nodes
git clone https://github.com/Madlumi/ComfyUiNodes

Then restart ComfyUI. No pip requirements or model downloads - pure Python on ComfyUI's own deps. Find it under Add Node → mnodes → int.

That's the whole review: if you already run the pack, this is a free primitive; if you don't, subtraction-as-a-node is a small reason to grab it alongside the rest. Just remember the 32-bit input clamp and you'll never have a surprise here.