KSampler Inputs

KSamplerInputs is a knob-puller, not a sampler. It doesn't denoise anything - it just lets you set the four dials every KSampler has (sampler, scheduler, CFG, steps) in one place and then feeds them to as many sampler nodes as you want. In a workflow with a main pass and a refine pass, or a txt2img + img2img pair, that's the difference between hunting through three identical nodes for "where did I put CFG 7" and changing one number.

The organizational win is the whole point, and it's real. AUN is a pack written by someone who builds dense, compact workflows, and this node is that philosophy distilled: one visible control instead of buried widgets.

How it works

It's a pure pass-through. The input dropdowns are pulled straight from ComfyUI's own sampler tables (comfy.samplers.KSampler.SAMPLERS and SCHEDULERS), plus a few extra schedulers the stock node doesn't list - the AYS presets (SDXL, SD1, SVD) and GITS[coeff=1.2]. Whatever you pick, it returns the four values unchanged. No compute, no randomness, no surprises.

The outputs are typed loosely (* for sampler and scheduler) so they'll plug into any node that accepts those inputs: a plain KSampler, KSamplerAdvanced, the AUN KSampler Plus nodes, whatever. The cfg output is a FLOAT and steps an INT.

The inputs that matter

• sampler - 44 options from ComfyUI's built-in list. euler, dpmpp_2m, the ancestral ones, and the newer *_cfg_pp family all show up.
• scheduler - 13 options including karras, sgm_uniform, exponential, and the AYS/GITS additions.
• cfg - FLOAT, default 2.0, range -2 to 100. Yes, it goes negative; that's for CFG++ / negative-guidance-style sampling, not a typo.
• steps - INT, default 10.

What the defaults say about you

Notice the defaults: CFG 2, steps 10. That's not a mistake - it's the modern bias. For SD 1.5/SDXL and their fine-tunes (Illustrious, Pony, NoobAI), you'll typically want CFG 4–7 and more steps; karras is your friend there. For distilled or flow-matching models (Turbo/Lightning checkpoints, newer 2026 releases), CFG 1–2 and low step counts are the norm, and aggressive schedulers like karras can actively hurt. The node's defaults lean toward the latter crowd - which is sensible, since CFG 7 burns a lot of modern models. The KB's CFG panel is a good read if you want the per-architecture sweet spots.

The gotcha

To wire these into a stock KSampler, you first have to convert that node's widgets to inputs: right-click the KSampler → Convert Widget to Input for sampler, scheduler, cfg, steps. Connect the four outputs, and from then on the widgets vanish from the KSampler and everything is controlled from your one node. If a connection just won't snap on, that conversion step is almost always the reason.

Installing it

Same story as every node in this pack. ComfyUI-Manager → search "AUN" → install, or:

cd custom_nodes
git clone https://github.com/loz2754/AUN-ComfyUI-Nodes

Restart ComfyUI. This particular node has zero heavy dependencies - it only reads ComfyUI's built-in sampler tables - but the pack's requirements.txt (piexif, opencv-python-headless, imageio-ffmpeg, requests) still gets installed for the image/video side, and Manager handles that automatically.

One honest note: this node shines in big multi-sampler graphs. For a single KSampler, it's just an extra hop for no gain. Use it where the centralization actually pays off.