Nodes/ComfyUI-Basic-Math/Precise Float | Basic
ComfyUI Node

Precise Float | Basic

A float widget with ten-decimal-place steps for micro-tuning

By akatz-ai·Created about a year ago·Updated 11 months ago· 16
Precise Float | Basic
    • FLOAT
    value0

    PreciseFloatInput ("Precise Float | Basic") is FloatInput's fussier twin. Same idea - one widget, one FLOAT out - but where the regular float input moves in steps of 0.001, this one steps at 0.0000000001 (ten decimal places). It exists for the values where a thousandth of a step is still too coarse.

    The single input is value: a FLOAT with the same ±1e12 range as FloatInput, but a step of 1e-10. The output is FLOAT.

    Why you'd reach for it

    Honestly? For most things you don't. A denoise strength at three decimals is more precision than the sampler can meaningfully use. But there's a small class of values where fine control genuinely matters: LoRA weights being nudged a fraction of a percent, tiny CFG tweaks where you're chasing the difference between 4.2 and 4.21, or calibration constants you're fitting by hand. When you find yourself fighting a slider that jumps by 0.001, this is the node that stops the fight.

    There's a second, quieter reason to use it: widget step is a contract. A FloatInput with step 0.001 makes it easy to type a clean three-decimal value and keeps the widget tidy for collaborators. PreciseFloatInput signals "this value is sensitive" to anyone reading the graph, and it preserves whatever precision you type instead of nudging toward a coarse grid.

    Installing it

    Same pack, same routine - pure Python, no dependencies, no models:

    cd ComfyUI/custom_nodes
    git clone https://github.com/akatz-ai/ComfyUI-Basic-Math
    

    Restart ComfyUI, find it under Basic Math ➕ → Primitives as "Precise Float | Basic". Or use ComfyUI Manager → **Install Custom Nodes → search "ComfyUI-Basic-Math"`.

    The gotchas

    The precision is in the step, not in the storage. It's still a regular Python float, so beyond about 15 significant digits you're back to floating-point reality - don't expect this node to magically carry 20-digit accuracy. And if you feed it into a chain that rounds early (NumberRound, for instance), all that fine precision gets flattened downstream, so keep the precise value on a path that preserves it. Also: "Precise Float | Basic" is the menu name, same | Basic suffix as the whole pack.

    CategoryBasic Math ➕/Primitives

    Inputs (1)

    NameTypeDefaultDescription
    valueFLOAT0-999999999999–999999999999

    Outputs (1)

    NameTypeDescription
    FLOATFLOAT