Min Float

Min Float does one thing: it takes two floats, a and b, and hands you back the smaller one. That's the entire node, and honestly the entire point of it. In the ComfyUI world of giant packs that need a PhD to install, a two-input min() you can wire in two seconds is refreshing.

Why you'd reach for it

You'll mostly use this to cap a value from above without pulling in a full clamp node. Say you're computing a denoise strength or a CFG from some upstream math, and you want to make sure it never exceeds a safe ceiling. Feed the computed value into a, feed your cap into b, and the output is whichever is lower. min(computed, 0.65) reads better in your head than three nodes of clamp plumbing, and it's one less place for a mistake to hide.

It's also handy whenever you're comparing two measurements that come from different places - two ControlNet strengths, two Float primitives from a scheduling setup - and you want the conservative pick. Equal values pass through unchanged, so there's no edge case where it surprises you.

How it works

Under the hood it's literally Python's built-in min(a, b) - no imports, no dependencies, no torch tensors, nothing to download. The author's description is "Returns the smaller of two floats," which is the whole contract. This is about as transparent as a node gets: you can see exactly what it'll do before you click queue.

The inputs that matter

Both inputs are required, both are FLOAT, and both default to 0:

• a - first float
• b - second float

That's it. One output, also FLOAT, which you can wire anywhere a float is expected - a KSampler's denoise, a CFG input, an upscale factor, whatever. It sits in the SimpleLogics/Math menu category, right next to its integer sibling Min Int.

Installing it

This is from the tiny danipisca07/ComfyUI-SimpleLogics pack (author Daniele Piscaglia), which ships no requirements file and no models - the whole thing is pure Python. Either search "SimpleLogics" in ComfyUI Manager, or clone it yourself:

cd ComfyUI/custom_nodes
git clone https://github.com/danipisca07/ComfyUI-SimpleLogics

Then restart ComfyUI. That's the entire install - there's no pip install, no weights to chase, which is a genuinely nice change from the dependency treadmill the rest of the ecosystem runs on.

Common issues

There isn't much to break here. One thing worth knowing: the repo also registers an unrelated "GeoCalib → fSpy JSON" camera-calibration node that the README never mentions, so don't be confused when a random Blender helper shows up in your node list after installing. Beyond that, the only realistic problem is a stale install - since it's a young, single-commit repo, git pull inside the folder (or Manager's update button) keeps it current. If you're getting a red node on something this simple, check your wiring before you suspect the node itself.