RandomInt

If you've ever wished ComfyUI's built-in seed node had a "stay inside this range" option, this is it. RandomInt (the display name of CM_IntRandomNumber) is a self-contained integer generator with four behaviors and six range constraints, and it's the newest addition to ComfyMath. Set it to Randomize, pick Custom min/max, and it'll hand you a random integer in your chosen range every run - no giant astronomically-sized seeds, no clamping shenanigans downstream.

It's genuinely useful beyond seeds, too. Need a random batch count between 2 and 8? An incremental frame counter for a video workflow that steps every run? A value that counts down and clamps at zero? RandomInt covers all of it with one dropdown, because it bundles the seed-style behavior (control_after_generation) together with range constraints in a way the stock seed node just doesn't.

The inputs that matter

• value - INT, the current value. It's the one you'll watch; the node writes generated results back to this widget.
• constraint - six options: Full range, 1 to max Int, 0 to max Int, Min int to -1, Min int to 0, Custom min/max.
• min / max - INT, default −1 and 1. Only used when Custom min/max is selected - they're always visible, and that trips people up. Set Custom min/max, then set your real bounds.
• control_after_generation - Fixed, Increment, Decrement, or Randomize (the default).

Output is a single INT. Everything uses signed 64-bit integer fields, so the range ceiling is roughly ±9.2 quintillion.

How it works, including the sneaky bits

When control_after_generation is Fixed, the node just returns value and tells ComfyUI nothing changed - so it doesn't rerun. Any other mode returns a deliberately "always changed" signal (the NaN trick the engine uses to force a node to run every single time), then rolls or steps the value. Increment and Decrement clamp to your bounds, so a counter can't walk off the edge. Under the hood it uses random.SystemRandom, the OS-level RNG, which is about as good as it gets for non-repeating randomness. A small frontend hook in the pack writes the generated value back into the visible value widget so what you see is what you got.

The trap everyone hits once: same as the classic seed control, the "after" in control_after_generation means the widget updates after the run - the number shown is the one that will be used next, not the one that just executed. If you randomize, see a great result, and switch to Fixed to keep it, you've already lost it. The stock ComfyUI advice applies: switch "widget control mode" to Before in your settings and this whole class of pain disappears.

Second gotcha: when the node isn't Fixed, it's permanently "dirty," which defeats ComfyUI's caching for everything downstream. In a workflow where RandomInt feeds a big sampler chain, that means no caching benefits - fine for occasional runs, mildly wasteful in a batch loop. That's the documented cost of "must run every time."

Installing it

It ships in ComfyMath, published as ComfyMath-NG. Install via ComfyUI Manager - search "ComfyMath" - or clone:

cd ComfyUI/custom_nodes
git clone https://github.com/rabanti-github/ComfyMath.git ComfyMath-NG

Restart ComfyUI. This fork registers the same node IDs as the original ComfyMath, so disable or remove the original first. Only dependency is numpy; nothing to download.

The one-line recommendation

For seeds: constraint 0 to max Int keeps you out of negative-seed weirdness, or Custom min/max for a tight band. For counters and ramps, Increment/Decrement with bounds is the cleanest loop logic you'll find in a single node.