Sigmas Hyperbolic

Sigmas Hyperbolic runs a noise schedule through one of six hyperbolic trig functions - sinh, cosh, tanh, and their inverses asinh, acosh, atanh - as a curve-reshaping transform. If you've done any machine learning, tanh (the default) is a familiar shape: an S-curve that saturates smoothly toward -1 and 1, and it's exactly the kind of squashing function that shows up all over neural network activations. Applied here, it's another entry in RES4LYF's "reshape the schedule with a named mathematical function" family, alongside Sigmas GammaBeta and Sigmas Gaussian.

What each function actually does to a curve

• tanh (default) - smooth S-curve, saturates at the extremes. Compresses large values much more than small ones.
• sinh - the opposite character: grows explosively for larger inputs, near-linear for small ones. Expands rather than compresses.
• cosh - always positive, flat near zero, grows steeply either direction - useful for turning a schedule symmetric around zero into something that's always positive.
• asinh, acosh, atanh - the inverses of the above three, so they undo the corresponding forward transform, or (applied fresh) invert which parts of the curve get compressed vs. expanded relative to their forward counterpart.

The inputs and outputs that matter

• sigmas (SIGMAS, required) - the schedule to reshape.
• function (enum: sinh, cosh, tanh, asinh, acosh, atanh; default tanh) - which function drives the transform.
• scale (default 1, range 0.01–10) - scales the input before the function is applied, which effectively controls how far into the function's curve (its flat regions vs. its steep middle) your schedule's values land.
• normalize_output (default true) - rescale the result back into a usable sigma range.

Output is a single SIGMAS list.

Should you use this?

tanh's saturating-S shape is at least conceptually similar to what a Gaussian CDF or a beta-distribution schedule does - compressing effort away from the extremes and toward a transition zone - so of the six options it's the one with the clearest connection to schedule shapes people already know work. That said, I found no community discussion of this node specifically, and no shared settings for any of the six functions. If you want a proven S-curve-shaped schedule, beta57 has actual track record behind it; this node is the unguided, more mathematically raw way to explore similar territory.

How to install it

• ComfyUI Manager - search "RES4LYF", install, restart.
• Manual - activate your venv, cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF, cd RES4LYF, pip install -r requirements.txt (portable builds: use the embedded pip.exe). Restart.

Common issues & troubleshooting

acosh is undefined below 1 and atanh is undefined outside (-1, 1). If your schedule (or its scaled version) has values outside those domains, expect NaNs. This is the most likely source of a broken output - switch back to tanh, sinh, or cosh (all defined everywhere) if you hit this, or adjust scale so your values land inside the valid domain.

A high scale pushes tanh/sinh/cosh into their extreme, flat-or-explosive regions where most of the schedule's original variation gets crushed or blown out. Start near scale = 1 and nudge it, checking the output with a preview each time.

Keep normalize_output on unless you're chaining directly into another sigma-math node that expects a raw, un-rescaled range - these functions can produce outputs far outside a normal sigma range depending on scale.