Sigma Waveform Node Simple

If the full Sigma Waveform Node is the lab bench, this is the hand tool. Sigma Waveform Node Simple - from the SigmaWaveFormNodes pack - does the exact same thing: generate a noise schedule by drawing a waveform. It just drops the fiddly extras and gives you a smaller set of shapes and settings, which honestly makes it the better node to learn the pack on.

It's aimed at the same niche as the whole pack: rolling your own SIGMAS schedule when the stock scheduler dropdown isn't enough. You want to see what a sawtooth distribution of denoising effort does to an image. This is the low-commitment way to find out.

What's different from the full node

• Five waveform types instead of seven - sine, sawtooth, stepped, square, random. No triangle, no rectangular.
• Tighter ranges - amplitude tops out at 10 (the full node goes to 100), frequency at 1.0.
• No Fourier toggle, no min_sigma floor, no damping switch. Notably, damping is always baked into the wave generators here - the exp(-damping_factor * t) decay is applied inside every shape, with no enable_damping boolean to turn it off. The full node's damping is optional; here it's just how the thing works.

The remaining inputs are the ones that matter: steps (default 200), amplitude, frequency, damping_factor, waveform_type, and enable_unidirectional. Keep unidirectional on unless you enjoy negative noise. One output, sigmas, into ComfyUI's native SamplerCustom node.

Installing it

Part of the SigmaWaveFormNodes pack, so one install gets all six nodes. Via ComfyUI Manager search "SigmaWaveFormNodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/BenNarum/SigmaWaveFormNode

Restart ComfyUI. Dependencies are numpy and torch - both already in any ComfyUI environment, and the pack downloads no models.

The catch to know about

Same fundamental warning as its big siblings: this produces a waveform, not a proper descending noise schedule. Even after the unidirectional shift it isn't sorted, so expect chaotic output - that's the experiment. And the random shape has no seed, so identical settings give you a different schedule every run. If you're A/B testing, use a deterministic shape like sine or sawtooth, or you'll be chasing a moving target.