Float Math

Float Math takes two floating-point numbers, applies an operation, and returns a float. It's the decimal twin of Int Math - same idea, but for the values a loop actually tunes: denoise, CFG, a scale factor, a quality threshold. If your Cyclist loop is nudging a number a little each pass, this is what does the nudging.

Where it fits

Cyclist loops by re-queuing the workflow, and Float Math is the arithmetic in the middle of a float loop: Recall Float → Float Math → Memorize Float. Ramp denoise down a touch each iteration, grow an upscale factor, or decay a threshold - all just "recall the value, do one op, store it back." The pack's docs also point out a sneaky-useful trick: because a Primitive node can't connect to an unspecified/wildcard input, you can use Float Math with an "add 0.0" to launder a float into something that will connect. It just passes the upper value straight through. Ugly, but it works when you're wiring into one of Cyclist's *-typed inputs.

The inputs that matter

• operation (enum) - one of seven: addition, subtraction, multiplication, division, exponentiation, max, min. Note this is real division (not floor division like the Int version) since floats don't round.
• float_1 (FLOAT, default 1, step 0.01) - first operand, the "upper" value.
• float_2 (FLOAT, default 1, step 0.01) - second operand.

Output is a single FLOAT. Order matters for the non-symmetric ops: float_1 - float_2, float_1 / float_2, float_1 ** float_2.

Install

• ComfyUI Manager: search comfyui-cyclist, install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/Pos13/comfyui-cyclist, restart.

Pure Python - no models, no dependencies.

Honest take

Like Int Math, this isn't a node you'd install Cyclist for - plenty of utility packs (pythongosssss Custom Scripts, WAS, rgthree) already do float arithmetic, often with more operations. Its value is being right there in the cyclist menu next to Recall/Memorize Float, so a value-ramping loop stays in one pack. Take it as a convenience that ships with the loop machinery, not a reason to install anything.

Watch the usual float things: division by zero errors, and floating-point comparisons are imprecise - which is exactly why Cyclist's Compare Anything node compares floats with a tiny tolerance rather than exact equality. If you're comparing the output of Float Math, use Compare Anything rather than an exact == somewhere else. And the pack-wide note: Cyclist is archived and unmaintained; this node is simple enough to be safe, but nobody's shipping fixes.