Sigmas Sigmoid

This is a sigma-shaping utility: give it an existing schedule and it runs the values through a sigmoid-family curve, redistributing where the denoising effort lands. It's one of the many "modulate the schedule versus time" tools RES4LYF ships, and like most of them it's a power-user knob - useful if you know what a sigma curve is and want to nudge its shape, ignorable otherwise.

The practical effect: an S-curve pushes the schedule to spend more steps at the extremes and glide through the middle (or the reverse, depending on the variant and gain). That changes the balance between structure and detail without you hand-typing every sigma.

How it works

It takes a SIGMAS input - a schedule you generated with some other scheduler node - and remaps each value through the chosen activation function. It's a transform, not a generator: you feed it a curve, it hands back a reshaped curve. That output goes to your sampler's sigmas input, overriding the sampler's built-in scheduler.

The inputs and outputs that matter

• sigmas (SIGMAS) - the schedule to reshape. You need to produce this upstream first (a scheduler node, or one of the other RES4LYF sigma nodes).
• variant (default logistic) - which curve. logistic and tanh are the classic smooth S-curves; the others (softsign, mish, swish, hardswish) are alternative activation shapes with subtly different bends. Start with logistic.
• gain (default 1) - how aggressive the curve is. Higher gain = sharper S, more redistribution. This is your main dial.
• offset (default 0) - shifts the curve's center, moving where the steep part falls.
• normalize_output (default true) - keeps the reshaped schedule anchored to sane endpoints. Leave it on; turning it off is how you accidentally make a schedule that doesn't fully denoise.

Output is SIGMAS (SIGMAS) → the sampler.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Manually:

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

Portable ComfyUI: embedded pip. Restart, hard-refresh F5. No downloads.

Common issues & troubleshooting

It errors or does nothing - I didn't feed it sigmas. This node reshapes an existing schedule; it doesn't invent one. If nothing's plugged into sigmas, there's nothing to transform. Put a scheduler node in front of it.

The image comes out under-denoised. Usually normalize_output got turned off, or a big gain bent the curve so it no longer lands near zero. Turn normalize back on and pull gain toward 1.

I can't see any difference. At gain 1 with offset 0 the reshape is gentle. Push gain up to actually see the effect, and change subject matter matters too - some prompts barely react to schedule shape.

Do I need this? No. It's a fine-tuning toy for people already deep in schedule-sculpting. If you're chasing quality on a flow-matching model and haven't tried beta57 yet, do that first - it's a much bigger lever than reshaping a curve by hand.