Sampler Settings

A small, single-purpose node: you choose a sampler and a scheduler, and it outputs those two choices as typed values for the rest of the Searge workflow to consume. That's it. No steps, no CFG, no model - just the sampler/scheduler pair, split out so the sampling node can wire them in cleanly.

It sits under Searge/_deprecated_/Inputs, so it's part of the older v2/v3 building-block set. In the current v4.x workflow the sampler choice is folded into SeargeGenerationParameters instead, but this node still resolves and still works if you're running an older graph or building one by hand.

How it works

ComfyUI treats the sampler (the algorithm that removes noise each step) and the scheduler (how much noise comes off each step) as two separate choices. This node exposes exactly those two and emits them on separate wires - SAMPLER_NAME and SCHEDULER_NAME - so a Searge sampling node can take them without you re-typing anything. It's a wiring convenience, not a piece of logic.

The inputs and outputs that matter

There are only two of each:

• sampler_name - 44 choices, default ddim. For SDXL the community default is DPM++ 2M Karras (dpmpp_2m paired with the karras scheduler). Euler a is fine for anime-style output but tends to over-smooth realism.
• scheduler - 9 choices, default ddim_uniform; you'll usually want karras alongside DPM++ 2M.

Outputs are the matching sampler_name and scheduler values, typed for downstream use.

The sampler/scheduler pairing is genuinely the choice that matters most here, and the good news for SDXL is that it's settled: SDXL and its finetunes (Juggernaut, RealVis, Pony, Illustrious) are DDPM-style models, and the old DPM++ 2M Karras default holds up on all of them. The "don't use Karras" advice you may have seen is about newer flow-matching models like Flux and Z-Image - it doesn't apply to SDXL. Don't carry that warning backward onto this pack.

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: run python -m pip install opencv-python in ComfyUI's Python environment first - it's a required dependency the pack won't work without - then cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git and restart ComfyUI. On Windows portable, the pack's installer script does all of this for you.

No models to download for this node; it only selects names.

Where people get burned

• Leaving it on ddim/ddim_uniform. They're valid but dated defaults. If your SDXL results feel soft or low-detail, switch to dpmpp_2m + karras before touching anything else.
• Importing flow-matching sampler advice. SDXL wants Karras. The blanket "Karras is broken" rule is a Flux/Z-Image thing; on this pack it's wrong.
• Reaching for it in a new workflow. It's deprecated. In the current v4.x setup, set the sampler inside SeargeGenerationParameters - you don't need this separate node anymore unless you're maintaining an old graph.