DP Float Stepper

Most float nodes hand you one static number. This one walks. You give it a start, an end, and a step size, and it steps through the range - the value advances each time you run, so a series of queued generations sweeps the parameter for you instead of you retyping it. That's the use case: "show me this at denoise 0.3, 0.35, 0.40, 0.45…" without babysitting the field between runs.

If you've ever wanted a quick, one-dimensional X/Y-plot without setting up a whole plotting node, this is the low-ceremony version. Point it at a strength, a CFG, a denoise - anything that takes a float - queue a batch, and you get an even sweep across the range you defined.

The inputs that matter

• start_point (−10 to 10, step 0.05, default 0) - where the sweep begins.
• end_point (−10 to 10, step 0.05, default 1) - where it stops. The negative range is there for parameters that legitimately go below zero; for a normal 0–1 sweep, leave start at 0 and set end to 1.
• step (0.01 to 1, default 0.05) - how far the value moves each step. Smaller step = more, finer samples across the range; bigger step = fewer, coarser ones.

Output is a single value (FLOAT) you wire into whatever parameter you're sweeping.

How to actually use it

Decide what you're testing - say CFG from 3 to 9 - set start_point 3, end_point 9, pick a step that gives you a sensible number of samples (step 1 gives you 3,4,5,6,7,8,9; step 0.5 doubles the resolution). Wire the output into the CFG field, then queue that many runs. Each one nudges the value along and you end up with a clean comparison strip. Pair it with a fixed seed so the only thing changing between images is the swept parameter - otherwise you can't tell what caused what.

How to install it

ComfyUI Manager: search ComfyUI-Desert-Pixel-Nodes, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/DesertPixelAi/ComfyUI-Desert-Pixel-Nodes

then restart. No models, no dependencies. Under the "DP" utils category.

Common issues & troubleshooting

The value isn't advancing. A stepper that walks a range needs successive executions to move - a single run gives a single value. Queue multiple runs (or a batch) to see it step. If it still won't move, something upstream is caching the node's output; make sure the graph is actually re-executing this node each time.

My comparison is noisy - I can't tell what the parameter did. Your seed is moving too. Lock the KSampler seed so the swept float is the only variable. A sweep is only readable when everything else is held constant.

I want it to stop cleanly at the end. Set start_point, end_point, and step so the range divides evenly - e.g., 0 to 1 by 0.05 is 21 clean samples. Odd combinations can leave you a partial final step. And remember the practical caps: values live in −10 to 10, step in 0.01 to 1, so very fine sweeps over a wide range take a lot of runs.