Scheduler Selector

Everybody obsesses over the sampler and forgets the scheduler, which is a shame because the two work as a pair. The sampler decides how noise gets removed each step; the scheduler decides how much. This node is the scheduler half of Helper-Nodes' one-choice-on-a-wire philosophy: pick karras once, and every node that needs to know - KSampler, Save Image - gets it from the same wire.

It's a boring node on purpose. That's the point.

How it works

Same passthrough trick as the sibling Sampler Selector. The dropdown reads comfy.samplers.KSampler.SCHEDULERS, the stock list, so what you see is exactly what the built-in KSampler offers: normal, karras, exponential, sgm_uniform, simple, ddim_uniform, beta, linear_quadratic, and kl_optimal. Nine choices, no additions, no hidden behavior. You select scheduler_name, the scheduler output echoes it back.

Inputs and outputs

• scheduler_name - the one required input, a 9-item enum.
• scheduler output - the same list as its type. Wire it into KSampler's scheduler and HelperNodes_SaveImage's scheduler input.

Nothing else exists on the node, so don't go hunting for options.

The honest take

The pair to remember: sampler + scheduler travel together, and the community even fuses them in names like "DPM++ 2M Karras." If you're on SD 1.5 or SDXL, karras has been the community default for years and is still a reasonable starting point. If you're on a flow-matching or distilled model, the aggressive schedules like karras and exponential can actively hurt - the model was trained on a near-straight trajectory and doesn't want a reshaped noise curve. When in doubt, take the pair your model card names, not the one your old habit does.

The real payoff here is the same as everywhere else in this pack: the choice becomes data. Save Image will bake the scheduler into your file's metadata, so when you're digging through 200 outputs a week later asking "which one was this?", the answer is already in the file.

Install

Search ComfyUI-Helper-Nodes in ComfyUI Manager and install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/teward/ComfyUI-Helper-Nodes

Restart after. Dependencies are numpy, pillow, piexif and whratio - numpy and pillow are already in ComfyUI, and Manager pulls the rest. No models involved.

Troubleshooting

Same rules as its sibling: missing node means restart, unfamiliar scheduler names mean a different ComfyUI version, and an unwired output does nothing. The node has no failure modes of its own - if it's installed and the dropdown populated, it works.