Sigmas Easing

Sigmas Easing takes an existing SIGMAS schedule and reshapes its curve using easing functions - the same cubic, quad, sine, expo, bounce family that animators and CSS folks know. In plain terms: it changes how the noise comes off over time without changing which scheduler you started from. Feed in a beta or simple schedule, ease it, and you've moved where the denoising effort concentrates. Its healthy click-through says people go looking for exactly this kind of curve control.

Why care about the curve shape? Because on flow-matching models the distribution of denoising effort across steps is one of the levers that actually pays. Where you concentrate the work - early for composition, late for detail - shifts the character of the output. Easing is a fast, intuitive way to nudge that without hand-building a schedule value by value.

How it works

You give it a schedule and pick an easing type and mode. The node normalizes the curve (optional), applies the easing math, and normalizes back out, giving you a modified SIGMAS that still spans the same range but bends differently in the middle. Because it operates on any incoming schedule, you can stack it after ClownScheduler or a stock scheduler and treat it as a curve post-processor.

The inputs that matter

• easing_type (default cubic) - the curve family: sine, quad, cubic, quart, quint, expo, circ, back, elastic, bounce. cubic and quad are the gentle, sensible ones; back, elastic, and bounce are aggressive and mostly for experimentation.
• easing_mode (default in_out) - whether the easing applies at the start (in), end (out), or both (in_out). This decides which end of the schedule gets reshaped.
• strength (default 1) - how much easing to apply. At 1 you get the full effect; lower blends toward the original curve.
• normalize_input / normalize_output (both default true) - keep these on unless you have a specific reason; they keep the eased curve inside the same sigma range so it stays valid for the sampler.

The single input schedule is sigmas (SIGMAS), and the single output is SIGMAS - plug it into a sampler's sigmas input, or into another sigma node.

Installing it

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

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

pip install inside the venv (or portable Python), restart, hard-refresh F5. No downloads.

Common issues

Two things. First, the exotic easings - elastic, bounce, back - can overshoot and produce a curve that dips or spikes in ways a sampler doesn't like, which shows up as artifacts. Keep normalize_output on, start with cubic/quad, and preview before you commit. Second, this is a curve modifier, not a scheduler - it needs a SIGMAS input to bend, so wire a scheduler into it first. And the golden rule for every sigma node in this pack: pair it with SigmasPreview so you can see what your easing actually did. Reshaping a schedule blind is how you end up with a mystery quality regression. If the node's missing, RES4LYF isn't installed.