basic_Ksampler_custom

ComfyUI has an entire family of atomic sampling nodes - separate NOISE, GUIDER, SAMPLER, and SIGMAS objects you assemble yourself with "SamplerCustomAdvanced" - for people who want to swap out one piece of the sampling algorithm (a custom guider, a hand-tuned sigma schedule) without touching the rest. basic_Ksampler_custom is that same fully-modular pattern, wrapped for Apt_Preset's context pipeline. Every field on this node is optional, which is the tell: it's designed to slot into a graph where most of the pieces already exist somewhere else, either as a context or as individually-wired atomic objects.

What it's actually for

This is not the node to reach for on a normal first pass - that's basic_Ksampler_full or basic_Ksampler_mid. This one is for when you're already deep enough into custom sampling that you're assembling NOISE, GUIDER, SAMPLER, and SIGMAS objects from other nodes (a custom CFG guider, a hand-built sigma curve, an alternate noise source) and you want the result to still speak this pack's RUN_CONTEXT language so it fits alongside the rest of an Apt_Preset workflow instead of breaking out into a completely separate wiring style.

The inputs and outputs that matter

Everything is optional: context, model, positive, negative, noise, guider, sampler, sigmas, latent, image, seed, denoise, and image_output (the shared Hide/Preview/Save/Hide/Save enum). Wire in as many or as few of the atomic sampling objects as you're customizing - anything you don't supply falls back to the context or an internal default. Outputs are the fullest of any node in this family: context, image, model, positive, negative, latent, and vae - everything a downstream stage might need, resolved and passed forward.

Installing it

Search "ComfyUI-Apt_Preset" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

Run install.bat on Windows (pip install -r requirements.txt on Linux/Mac), restart. No extra dependencies of its own - whatever custom sampler/guider/sigma nodes you're pairing it with will have their own requirements to check separately.

Common issues

Because nearly every field is optional, the most common problem is a quiet one: you meant to override the sampler or sigmas for this run, forgot to actually wire it, and the node silently fell back to the context's defaults instead of erroring - so your "custom" run behaves identically to a normal one and you spend time debugging the wrong thing. Before assuming a custom guider or sigma schedule isn't working, double-check every atomic input you meant to override is actually connected, not just present in the node's UI. This is also the node where getting the sampler/scheduler pairing wrong for your model architecture (Karras on a flow-matching model, for instance) does the most damage, since you're the one now responsible for that choice instead of a preset making it for you.