Sigmas TimeStep

Another entry in RES4LYF's deep bench of sigma-manipulation nodes. Sigmas TimeStep takes a noise schedule and reworks it against a timestep step-size and a decay, giving you another way to redistribute where the sampler spends its effort. It's squarely in "modulate parameters versus time" territory - the part of the pack the README says can pull real gains out of sampling if you're willing to fiddle.

Let's be honest up front: this is niche. If you're new to RES4LYF, you can build great images without ever touching a single sigma node. This one is for people who already think in schedules and want a specific kind of reshape.

How it works

It consumes a SIGMAS schedule you produced upstream and transforms it: dt sets an effective step size, scaling bends the relationship between timestep and sigma, and decay optionally tapers the values over the run. The result is a new SIGMAS schedule that plugs into your sampler and overrides whatever scheduler and step count you set inside the sampler node.

The inputs and outputs that matter

• sigmas (SIGMAS) - the schedule to transform. Generate it first with a scheduler node; this one doesn't create a schedule from nothing.
• dt (default 0.1) - the timestep step size the transform works against. Smaller values make finer adjustments.
• scaling (default linear) - the curve mapping timestep to sigma: linear, quadratic, sqrt, or log. linear is the neutral starting point; the others weight the transform toward the early or late part of the run.
• decay (default 0) - a taper applied across the schedule. 0 means no decay; raise it to progressively pull sigmas down over the run.

Output is SIGMAS (SIGMAS) → your sampler.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF
cd RES4LYF
pip install -r requirements.txt

Portable ComfyUI: swap in the embedded pip. Restart, hard-refresh with F5. Nothing to download.

Common issues & troubleshooting

Nothing to transform. Like the pack's other sigma-shaping nodes, this reshapes an existing schedule. If sigmas isn't connected to a real scheduler upstream, it has nothing to work on. Wire a scheduler node in first.

Residual noise in the output. Aggressive scaling or a heavy decay can leave the schedule not fully reaching the low end, so the image ends up under-denoised. Back the settings toward linear / 0 and reintroduce changes one at a time.

I can't predict the result. That's normal for a low-level schedule transform - it's a taste-and-iterate node, not a preset. If you have a sigma-preview/plot node, use it to actually see the curve. Otherwise, change one parameter, render, compare.

Is this the tool I want? Probably not, if you're picking your first RES4LYF settings. The biggest, easiest wins on flow-matching models come from a good sampler and beta57, not from hand-transforming sigmas. Come back to this node when a preset schedule genuinely isn't giving you the distribution you need.