Constant Scheduler

This is the simplest scheduler in RES4LYF, and it's honestly more of a building block than something you'd wire straight into a sampler for a normal generation. It produces a SIGMAS curve that goes from a start value to an end value in a plain, unshaped ramp - no beta curve, no Karras concentration, just a straight line (or a flat hold, if start and end match). Where a normal scheduler carefully redistributes denoising effort across the steps, this one deliberately doesn't.

So why does it exist? Because a lot of RES4LYF's power comes from feeding SIGMAS curves into places that aren't the main sampler: guide-weight schedules, sigma-manipulation chains, custom multi-stage setups. A constant or linear ramp is often the cleanest starting point for those. It's the "give me a predictable baseline curve I can then bend" node.

How it works

You set a step count and two endpoints, and it interpolates a straight line between them, with an optional cutoff so the curve stops early. That's it - no model-derived sigmas, no schedule math. The predictability is the feature.

The inputs and output

• steps (default 40) - how many values in the curve.
• value_start (default 1) and value_end (default 0) - the endpoints. Leave them at 1→0 for a descending ramp; set them equal for a genuinely constant schedule; reverse them for an ascending one.
• cutoff_percent (default 1) - where along the curve to stop, as a fraction. At 1 the full ramp runs; lower it to truncate.

The single output is sigmas (SIGMAS). Feed it into a sampler's sigmas input if you really want a linear schedule, or - more usefully - into a guide-weight or sigma-processing input that expects a curve.

Installing it

ComfyUI Manager: search RES4LYF, install, restart. Manually:

cd custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt

pip install inside the venv (or portable Python), restart, hard-refresh F5. No downloads.

Common issues

Don't use this as your main image scheduler and expect good results - a flat or purely linear sigma schedule isn't what a flow-matching model wants for a full generation, which is exactly why the community's advice is to take beta57, simple, or beta for actual image sampling. On a straight trajectory those balanced schedules are what perform; an unshaped ramp is a tool for building custom curves, not a quality setting. If your image looks flat or muddy after wiring this into a sampler, that's the likely cause - swap in ClownScheduler on beta57 instead. Pair it with SigmasPreview to see the line you're actually producing. If the node's missing, RES4LYF isn't installed.