🐐 Sampler Settings

If you've built more than a couple of ComfyUI graphs, you've hit this annoyance: steps, CFG, and denoise are three separate widgets living directly on your KSampler node, which means every time you want to reuse the same values elsewhere - a second sampler in a refine pass, a display node showing your settings, a save-metadata node - you're either retyping the same numbers or wiring up three separate primitive nodes to carry them. Sampler Settings is the fix: one node, three inputs, three matching outputs, so your generation settings become something you can pipe around the graph instead of something trapped on one widget.

How it works

There's no logic here beyond validation - it's a settings container. You dial in steps, CFG, and denoise once, and the node hands them back out as typed values (INT, FLOAT, FLOAT) that any node expecting those types can consume. The value in that is entirely about graph hygiene: keep one Sampler Settings node as your single source of truth, and every sampler, refiner, or logging node downstream reads from it instead of duplicating the numbers by hand.

Worth knowing what these three actually do if you're newer to this: steps is how many denoising iterations the sampler runs - more steps generally means more refined output up to a point of diminishing returns, and the sweet spot depends heavily on your sampler and scheduler. CFG (classifier-free guidance) controls how hard the model is pushed to follow your prompt versus wandering on its own - but if you're on a modern guidance-distilled model (Flux, Z-Image, and most 2026-era checkpoints), CFG 1 is correct and higher values just double render time while over-cooking the image; that rule flips for older SD1.5/SDXL-style checkpoints, which still want CFG in the 4–7 range. Denoise matters most when you're not starting from pure noise - a second pass over an already-generated image, an img2img step, or a hi-res fix - where 1.0 means "regenerate from scratch" and lower values (0.3–0.5 is the common range) mean "keep the composition, just add detail."

Inputs and outputs that matter

• steps (0–100, default 20)
• cfg (0–100, default 4, step 0.5)
• denoise (0–1, default 1)

Every input has a matching output of the same name and type - steps (INT), cfg (FLOAT), denoise (FLOAT) - wire them straight into your KSampler's corresponding widgets (right-click → convert to input first if they're not already inputs on your sampler).

How to install it

Via ComfyUI Manager: search "GOAT Nodes", install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/AconexOfficial/ComfyUI_GOAT_Nodes.git

then restart ComfyUI. Nothing to download, no dependencies beyond the pack itself - this is one of the lightest nodes in the set.

Common issues & troubleshooting

The default CFG of 4 is wrong for your model, and that's not the node's fault. This is a generic passthrough node - it has no idea what checkpoint you're running. If you're on a distilled model expecting CFG 1 and you leave the default 4 in place, you'll get an overcooked, burned-looking image; that's the model reacting to CFG, not a bug in Sampler Settings. Set the value for the model you're actually using.

Nothing changes when you adjust it. Almost always a wiring issue - make sure the sampler's steps/cfg/denoise widgets have actually been converted to inputs and connected to this node's outputs, rather than the node just sitting there unconnected next to a sampler that's still reading its own local widget values.

Sitting alongside a Triple Prompt / Image Dimensions node is the intended pattern here - GOAT Nodes' whole "Input" category exists to turn scattered widgets into reusable, wireable blocks, so it's worth pairing this with those two if you're building a template workflow you'll reuse often.