Collection Primitive (scheduler)

The scheduler-flavored twin of this pack's sampler node, and the same reasoning applies. A KSampler's scheduler widget, once converted to input, isn't a plain string - it's typed to the exact list of valid scheduler names, and that combo type doesn't pass through a Reroute node any better than the dynamic type on a native primitive does. This node gives you a scheduler dropdown with the type nailed down in advance, so you can set it once and route the value across your graph without ComfyUI blocking the wire.

What a scheduler actually is, quickly

If the sampler decides the algorithm for removing noise at each step, the scheduler decides how much noise gets removed at each step - the shape of the curve, not the method walking it. karras concentrates the work in the middle steps; normal spreads it evenly; and so on. They're independent choices - you pick both a sampler and a scheduler, and the pack ships a separate collection primitive for each, matching that split.

Inputs and outputs

One field: Input_scheduler, a dropdown with 6 choices - normal, karras, exponential, sgm_uniform, simple, ddim_uniform. Pick one and it's passed through as the single output, typed to that same scheduler-name list, ready to plug into anything expecting a scheduler - most commonly a KSampler's scheduler input once converted, or an advanced/custom sampling node with the same combo type.

The gap worth knowing about

Those 6 options are ComfyUI's original scheduler set, and they're the right call for SD 1.5, SDXL, and most Flux-era workflows - karras for the DPM++ family, normal as the safe fallback. But if you're on a flow-matching model, this list is missing exactly the two schedulers that current guidance actually recommends: beta and linear_quadratic, both called out for Z-Image Turbo specifically. Worse, karras and exponential - two of the six choices this node does offer - are actively the wrong pick on those newer models; they reshape the denoising curve in a way flow-matching architectures don't want. So on an SD 1.5/SDXL/Flux workflow, this dropdown has everything you need. On Z-Image, Anima, or similar, it's missing your actual target and still tempting you with two options you shouldn't use. If that's your situation, leave the target node's scheduler widget as a widget instead of converting it to input, and pick beta or linear_quadratic directly there.

Installing it

ComfyUI Manager - search ComfyUI-Static-Primitives, install, restart.

Manually:

cd ComfyUI/custom_nodes && git clone https://github.com/80sVectorz/ComfyUI-Static-Primitives.git

Restart ComfyUI. No dependencies, no model files, nothing else to configure. Find it under the primitives category alongside the rest of the pack.

Where people get tripped up

Picking karras out of habit on a newer model. It's the default reflex from years of SDXL workflows, and it's in this dropdown, but on Z-Image, Anima, and other flow-matching checkpoints it actively hurts results rather than helping. Check what your specific model card recommends before defaulting to it.

The scheduler you need isn't listed. As above - this is a fixed set of 6, not a live pull of everything registered on your ComfyUI install. beta and linear_quadratic aren't in it; neither is anything a third-party pack like RES4LYF adds, even if that pack is installed alongside this one.

Feeding the output into a plain STRING socket. It won't connect. This node's output is typed to the scheduler combo list specifically, not a generic string, even though the values look like plain text on the surface.

Mismatching it with the sampler node. Sampler and scheduler are independent settings that both matter - picking the right scheduler for the wrong sampler (or vice versa) won't throw an error, it'll just give you a worse image. Cross-check both against whatever the model's documentation recommends, not just one in isolation.