Dynamic Scheduler Selector

The sibling to this pack's Dynamic Sampler Selector, and built for the same job: assembling a set of values to sweep through in a comparison run, because ComfyUI has no native equivalent of A1111's XYZ plot script. The README says these generator nodes were "originally crafted to support XYZ-plot pipelines," and the community corpus backs that framing up - people regularly ask how to do grid comparisons in ComfyUI and the honest answer is always "possible, but you're building it yourself out of loops."

What it does

Ten dropdowns, scheduler_1 through scheduler_10, each defaulting to none and offering the full ComfyUI scheduler list: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, normal, linear_quadratic, kl_optimal. Pick the ones you want to compare, leave the rest on none, and the node filters those out and returns your selections two ways: a comma-separated STRING for labeling, and a LIST for actually driving a loop.

Same wiring pattern as its sampler counterpart: neither output plugs directly into a KSampler's scheduler slot, because that's a combo-typed connection and these outputs are generic data. Feed the LIST into a loop, and inside it use the pack's Scheduler Selector From String node to convert each entry back into a real scheduler type per iteration.

Installing it

ComfyUI Manager: search DemonAlone-nodes-ComfyUI, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/DemonAlone/SimpeStringGenerator_ComfyUI

No extra dependencies or model downloads in the README - it's a short credits list crediting a handful of better-known packs for pieces of this one. Small, low-traffic personal pack; don't expect a support community around it specifically.

Where people get burned

Leave all ten on none and you get empty outputs - nothing to sweep. But the bigger trap here is scheduler choice itself, and it's worth knowing before you build a ten-item comparison list: karras and exponential are safe defaults on SD 1.5 / SDXL and active failures on flow-matching models (Flux, Z-Image, and similar). The community's own explanation is that flow-matching architectures want a balanced sigma schedule and fall apart under aggressive redistribution of denoising effort - the kind karras and exponential apply. If you're building a scheduler sweep against a modern flow-matching checkpoint, beta, simple, and linear_quadratic are where the real signal is; including karras/exponential just wastes two slots confirming they don't work, unless that's exactly the point of your test.

One more landmine if you're pairing this with the pack's ShiftSliderNode or any timestep-shift setting: some schedulers - bong_tangent among them, from the RES4LYF pack - apply their own internal sigma shift and ignore whatever shift value your workflow sets. If shift seems to do nothing while you're sweeping schedulers, check which scheduler is active before concluding the setting is broken.