LTFloat_Step_00002

LTFloat_Step_00002 is one of the Latent Tools pack's step helpers: a float constant whose widget increments by 0.0002, with the value handed out as both a FLOAT and a STRING. The siblings go from 0.0001 to 1.0, and this one sits near the fine end - a 0.0002 step per image means it takes 500 runs to move a parameter by a tenth. That's a tool for patience and precision, not for quick looks.

Why would you want that? Because the pack's whole trick is that small changes to the starting noise's mean or standard deviation produce visible changes in the final image. When you're sweeping σ across a hundred frames, 0.0002 per frame is a smooth gradient; anything bigger starts reading as steps. The README's own noise sweeps run in the 0.001 range, so 0.0002 is for when you want an even longer, gentler ramp - or a frame-by-frame video noise drift that never visibly jumps.

How it works

Straightforward, and worth being honest about: the node does not increment by itself. You type a value, it returns that value rounded to five decimals as a FLOAT and a STRING. The "step" is the spinner's increment and the label that tells you what this constant is for. The incrementing is on you - bump the value between runs, keep everything else fixed.

The inputs and outputs

• value - FLOAT, default 0, range ±1,000,000, step 0.0002.
• FLOAT - the value.
• STRING - the value as text, ready for a filename prefix so each generated image records the parameter that produced it.

The pair is the point: FLOAT into a mean/std input on LTGaussianLatent, STRING into the filename of a Save Image node, and you've got a self-documenting sweep.

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.