LC Basic Scheduler

ComfyUI's built-in KSampler hides its sigma schedule from you, and that's fine for 95% of work. But the moment you want two different noise schedules running in sequence - a Karras pass into a beta pass, a low-step refine, a dual-model setup - you need the schedule as an object you can split. LC Basic Scheduler is the plumbing that produces it: give it a model, a scheduler name, and a step count, and it hands you a full SIGMAS curve with no denoise trimming.

The mechanism is exactly what the name says and not a single thing more. It calls ComfyUI's own sampler machinery - the same KSampler.SCHEDULERS list ComfyUI itself uses - and builds the full sigma curve for your chosen scheduler over the full steps count. Note the "full" part: there's no denoise parameter here, and that's deliberate. The node's description is blunt that this uses the entire step count with no denoise cut. If you want a partial schedule (say, the last 40% of steps for a refine pass), that's a different node's job - this one exists specifically so you can feed two complete curves into the pack's LC Split Sigmas (Advanced), which does the splitting for you.

Inputs are three and no more. model defines the sigma range - it has to match the model you're actually sampling, because the schedule's start and end values come from that model's noise space. scheduler is the dropdown with the usual suspects: karras, beta, exponential, simple, ddim_uniform, sgm_uniform, and more - 9 choices, same list as the KSampler. steps is the count (1 to 10,000, default 20). Output is one SIGMAS wire, which plugs into anything that consumes a sigma schedule: samplers, the split-sigmas nodes, or advanced sampler configurations.

The natural pairing is the one the description calls out: two of these, each with a different scheduler or step count, fed into LC Split Sigmas (Advanced), which merges them into a single dual-schedule run. That's the "dual-scheduler dual-pass" pattern the pack is pushing, and it's a genuinely useful technique when your base model likes Karras but your detail pass wants something tamer.

Where people get confused: the absence of a denoise knob. If you're used to the KSampler's denoise-as-fraction-of-steps mental model, this node's "full step count always" can surprise you - it's not forgetting the option, it's a design decision so the split node has complete curves to work with. And don't wire a sigmas output into a standard KSampler expecting it to be used - the built-in KSampler derives its own schedule; you need a sampler that accepts an explicit SIGMAS input (the advanced sampler, or this pack's sampler-configure family) for the custom curve to mean anything.

It's a small, focused node and the pack keeps it that way - no denoise, no extra outputs, just a schedule you can inspect and split. Install is the one clone:

cd ComfyUI/custom_nodes
git clone https://github.com/lonecatone23/ComfyUI_LC123_nodes

Restart, no pip deps. If you never split schedules, you'll never touch it; if you do dual-pass work, it's the clean way to build the two halves.