LTFloat_Step_0002

LTFloat_Step_0002 is the Latent Tools step helper with a 0.002 increment - the "fine but faster than the finest" tier of the family that runs from 0.0001 to 1.0. It's a calibrated constant: you give it a value, it hands back the same number as a FLOAT and a STRING, and the step size tells you how it's meant to be used.

Every node in this family exists to sweep a parameter across a batch of images with the seed locked. 0.002 per run is a good middle-fine choice: it still looks smooth when you chain the results, but a 100-image run moves your parameter by 0.2 instead of a tenth. If you know you want the README-style σ sweep but the 0.001 node feels a touch slow for your batch size, this is the sensible step up.

How it works

The honest model: the node does not auto-increment. It rounds the value you type to five decimals and returns it twice - once as a FLOAT, once as a STRING. The "step" is the widget's spinner increment and the label for the constant. The incrementing happens when you bump the value between runs; the node just makes that deliberate, visible, and labeled.

The inputs and outputs

• value - FLOAT, default 0, range ±1,000,000, step 0.002.
• FLOAT - the value.
• STRING - the value as text, which is the trick that makes sweeps self-documenting: paste it into a filename prefix and every output records the parameter that created it.

Typical wiring: FLOAT into the mean/std input of LTGaussianLatent, STRING into a Save Image filename. Fixed seed, nudge the value, compare.

Installing it

Part of xl0's Latent Tools pack. ComfyUI Manager → search Latent Tools → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/xl0/latent-tools

Restart, and it's under LatentTools. The pack's only dependency is lovely-tensors (Manager installs it; manual cloners may need pip install lovely-tensors). No model downloads.