Sigmas Attractor

This is one of the genuinely strange corners of RES4LYF, and I mean that as description, not complaint. Sigmas Attractor runs your noise schedule through one of five classic strange-attractor systems - the same chaotic differential equations you've probably seen visualized as swirling butterfly-wing plots - and reads a new sigma schedule back out.

What a strange attractor even is here

Lorenz, Rössler, Aizawa, Chen, and Thomas are all small systems of differential equations, famous in chaos theory for producing trajectories that are fully deterministic but never repeat and never settle - tiny changes in starting conditions produce wildly different paths over time, while the overall shape stays structured rather than random. Applying that to a sigma schedule is exactly as unusual as it sounds: instead of picking a smooth, well-understood curve like Karras or beta, you're driving your noise levels through a chaotic dynamical system and taking whatever comes out.

How it works

For iterations steps of size dt, the node steps the chosen attractor system forward, then reads off one axis - component: x, y, z, or the vector magnitude - as your sigma values, with normalize_output rescaling the (often wildly-scaled) chaotic trajectory back into something usable.

The inputs and outputs that matter

• sigmas (SIGMAS, required) - the input schedule.
• attractor (enum: lorenz, rossler, aizawa, chen, thomas; default lorenz) - which system to iterate.
• iterations (default 5, up to 50) and dt (default 0.01) - how many steps the system runs and how big each one is.
• component (enum: x, y, z, magnitude; default x) - which axis of the trajectory becomes your sigma output.
• normalize_output (default true) - rescale the result into a usable range.

Output is a single SIGMAS list.

Should you actually use this?

Honestly - probably not for a render you care about. I looked for any community report of someone using RES4LYF's chaos-system sigma nodes in a real workflow and found nothing; even in threads where people are actively hand-tuning custom sigmas (splitting schedules, rescaling them, sharing tweaked values for specific models), nobody's reaching for the attractor generators. This reads as one of the pack's more art-project-flavored inclusions - chaotic systems are deterministic but wildly sensitive to small changes, which could be interesting if you want reproducible-but-varied schedules across a batch, but there's no evidence it beats a tuned scheduler on actual image quality.

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

Output isn't guaranteed monotonic or non-negative. Chaotic trajectories don't respect "should decrease smoothly toward zero" - run the result through Sigmas Abs and Sigmas Cleanup, and check it with Sigmas Count or a preview, before it ever reaches an actual sampler.

Small dt or iterations changes can swing results wildly - that's the defining property of chaos, not a bug, but it means there's no stable "good setting" to hand you. If you're experimenting, change one parameter at a time and actually look at the output values.

No pack or community documentation exists for this node. You're on your own here more than with almost anything else in RES4LYF.