LTFloat_Step_00001

LTFloat_Step_00001 outputs the float you type into it, plus the same number as a string - and the value widget's increment is calibrated to 0.0001. It's the finest member of the Latent Tools pack's batch of step helpers, which run from 0.0001 all the way up to 1.0, one node per step size.

These nodes exist for one job: sweeping a parameter across a batch of images in controlled increments. You keep the seed fixed, nudge the value up by the step between runs, and watch how the output responds - the discipline of changing one variable at a time, mechanized. With a 0.0001 step you can make the change between consecutive frames almost imperceptible, which is exactly what you want when the goal is a smooth animation rather than a jump cut.

How it works

Here's the thing to know so you're not confused later: the node does not auto-increment on its own. It takes the value you give it and returns it rounded to five decimals. The step in the name is the widget's increment - how much the number jumps when you click the spinner - and the design intent is that you bump the value by that step between runs. It's a fixed constant, tuned and labeled, not a stateful counter.

The inputs and outputs

• value - a FLOAT, default 0, range ±1,000,000, stepping by 0.0001.
• FLOAT output - the value, rounded to five decimals.
• STRING output - the same value as text, which is the underrated part: it drops straight into a filename prefix, so every image in your sweep is labeled with the exact parameter that made it.

Wire the FLOAT into a mean or std input on LTGaussianLatent, and the STRING into a save-image filename prefix. For video noise sweeps, feed the float into a batch of per-frame noise generators and this is how you get the slow, deliberate drift that the pack's demo GIFs are made of.

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 - it's a constant and a string conversion.