Float 🐲

This is a float you set by typing, capped between 0 and 1, that outputs the value rounded to three decimal places. It sounds like the least interesting node imaginable. Then you spend an hour wondering why a 0.3 denoise in one node produces a different image than 0.3 in another, because one of them is secretly 0.30000000000000004. The rounding is the whole trick: it kills the floating-point drift that makes identical-looking values behave differently.

What it does

A single number widget, range 0.0 to 1.0, step 0.005, default 0.765. Output is a FLOAT rounded to three decimals. That's it. No slider - this is the "type a value" version, while its sibling Float Slider gives you the drag handle over the same range.

The range is deliberate. 0 to 1 is the natural range for denoise, the img2img strength knob, and any "how much of this thing" strength parameter. The pack's number nodes are color-coded by job, and the README is explicit about it: Float and Float Slider are for denoise, Float L is for CFG, Int Slider is for steps. Reach for this when you want a denoise value as a stable, inspectable number rather than a slider you can knock off-center.

The inputs

• number - the float. Min 0, max 1, step 0.005, default 0.765.
• Output: a single FLOAT.

Wire it to the denoise input on a KSampler (or an img2img pipeline) and you've got an explicit, repeatable value. Because it rounds, saving and reloading the workflow keeps the value byte-identical instead of drifting by a microtenth.

Installing it

Part of the RyuuNoodles pack:

cd ComfyUI/custom_nodes
git clone https://github.com/DraconicDragon/ComfyUI-RyuuNoodles

Restart ComfyUI, or install via Manager (search "RyuuNoodles"). No models, no extra dependencies.

Gotchas

The limits are hardcoded and the author has said they probably won't become configurable: you cannot get a value above 1.0 from this node, period. If you need a float above 1 (CFG scale, upscale factors), that's what Float L (0-24) is for. And because it's a plain number input rather than a slider, it's slightly less convenient for rapid tweaking - but for a value you want to be exact, that's a feature, not a bug.