Sigmas ConwaySequence

Almost every other node in RES4LYF's sigma toolbox transforms an existing schedule. Sigmas ConwaySequence is different - it doesn't take a sigmas input at all. It generates a schedule from scratch, built out of one of four classic integer sequences from combinatorics.

The four sequences on offer

• look_and_say - the "read the previous line aloud" sequence: start with 1, the next term describes it ("one 1") as 11, the next describes that ("two 1s") as 21, then 1211, 111221, and so on. Its terms grow enormous fast.
• audioactive - a close relative of look-and-say, same "describe the previous term" family.
• paperfolding - the sequence generated by repeatedly folding a strip of paper in half and reading off the pattern of resulting creases.
• thue_morse - a well-known sequence in combinatorics on words, famous for avoiding certain repeated patterns (cubes), and one that turns up in fields as unrelated as fair-division scheduling.

These aren't arbitrary - each one is a real, studied object in mathematics with genuine structure, not randomness. That's the whole appeal of using one here: a deterministic, non-random way to generate a schedule that isn't just a smooth curve either.

How it works

steps sets how many terms of the chosen sequence_type to generate. Because raw sequence values (especially look-and-say and audioactive) can grow to enormous or wildly uneven magnitudes, normalize_range squashes the result into a sane numeric window bounded by min_value and max_value.

The inputs and outputs that matter

• steps (default 20, range 1–50) - how many terms to generate.
• sequence_type (enum: look_and_say, audioactive, paperfolding, thue_morse; default look_and_say) - which sequence.
• normalize_range (default true) - rescale into min_value–max_value.
• min_value (default 0.01) / max_value (default 10) - the target output range.

Output is a single SIGMAS list. No sigmas input required - this node is a generator, not a transform.

Should you actually use this?

Same territory as the pack's chaos and attractor generators: I found zero evidence of anyone using RES4LYF's novelty sequence generators in a real workflow, and the README doesn't cover it beyond framing the whole sigma-manipulation category as experimental. It's a genuinely interesting curiosity - these sequences have real mathematical structure that random noise doesn't - but there's no established recipe for turning that structure into a better image. Treat it as something to explore, not something to reach for under deadline.

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

Check steps carefully. With look_and_say or audioactive, term lengths explode quickly - the underlying sequence values behind higher steps counts can be enormous even before normalization, so keep an eye on generation time and on whether normalize_range is actually doing sensible work at higher step counts.

Always verify the output before a full render - wire in Sigmas Count to confirm the schedule length matches steps, and preview the actual values. There's no monotonicity guarantee here any more than with the pack's other generators, since these sequences weren't designed with "decreasing noise schedule" in mind.