Scheduler Selector

People obsess over samplers and forget the scheduler, but "DPM++ 2M Karras" is only half a sentence without the "Karras." The scheduler decides how much noise gets removed at each step - the shape of the denoising curve. The Scheduler Selector is the sibling to the Sampler Selector in this pack: one dropdown that feeds the same scheduler name to every KSampler in your graph, so you stop hunting for the right widget each time you swap models.

That last bit is the real reason to install it. Different model families want different schedulers, and they disagree hard. SD 1.5 and SDXL love Karras. Flux hates it. Flow-matching models like Z-Image and Klein actively dislike Karras and exponential, preferring balanced schedules like simple, beta, or linear_quadratic. If you're running a workflow that switches checkpoints, a single shared scheduler widget means the difference between remembering all of that and retyping it each time.

How it works

Mechanically it's the same trick as the Sampler Selector: the node pulls its options live from comfy.samplers.KSampler.SCHEDULERS, so the dropdown always matches what your installed ComfyUI actually supports. Right now that's nine: normal, karras, exponential, sgm_uniform, simple, ddim_uniform, beta, linear_quadratic, and kl_optimal. It returns the string you picked, and you wire it into the scheduler input of a KSampler.

The one input is scheduler, default normal. The single output, also scheduler, carries the name. That's the whole node - the value is in the wiring, not the code.

Wiring it

Same drill as its sibling: right-click the KSampler's scheduler widget and pick Convert scheduler to input, then drag the output onto the new port. Repeat for every KSampler you want synced. Once it's connected, model swap is a two-click change instead of a per-node hunt. Connect the Sampler Selector the same way and you've effectively built a central "sampler + scheduler" control room for the whole workflow.

The honest caveats

This node only knows stock ComfyUI schedulers, which matters more than you'd think. The scheduler world moved: packs like RES4LYF ship their own schedulers - beta57 and friends - that show up in high-scoring 2026 workflows for Anima, Qwen-Image and Wan. If a workflow you downloaded uses one of those, this node won't list it; you'll need the pack that provides it. This is a "keep the main graph tidy" node, not a replacement for the sampler-tuning pack ecosystem.

Also: picking a scheduler is only half the advice. The rule of thumb that survives everything - Karras for SD 1.5/SDXL, never Karras on Flux (stick to normal or beta there), sgm_uniform for most distilled Turbo/Lightning models - still assumes you pair it with a sensible sampler. The Selector won't stop you from building "euler + karras" on Flux and wondering why it's wrong.

Install

ComfyUI Manager → search "ComfyUI_Selectors" → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ComfyAssets/ComfyUI_Selectors.git

Then restart ComfyUI fully. No models to download, no runtime dependencies to fight - the repo's requirements-dev.txt is dev-only tooling, so install is clone-and-go. The node appears under comfyassets/Sampling. If the output won't connect, you forgot to convert the destination widget to an input - that's the one thing that trips everyone.