Sampler Loop (Sampler Custom Advanced)

If you've already met the plain Sampler Loop from this pack, Sampler Loop (Sampler Custom Advanced) is that node with one difference, and the difference is the whole point: the sampler output is a real SAMPLER object, not a string name.

Why care? Stock KSampler is happy with "dpmpp_2m" as text. Custom sampler nodes - SamplerCustom, Kijai's SamplerCustomAdvanced, and the rest of the advanced-sampling crowd - want an actual callable object on their sampler input. Feed them a string and you'll get a type mismatch that reads like "argument 'sampler' expected type SAMPLER but got STRING" and zero output. This node exists so you can loop through samplers and feed a custom node at the same time.

How it works

Same counter machinery as the rest of the pack: it picks one sampler per workflow execution (advancing only on new runs, so a single queue run returns a stable value), then resolves the chosen name to an object via ComfyUI's comfy.samplers.sampler_object(). If that resolution fails it tries the internal samplers dict, and only as a last resort falls back to the raw string with a warning - which is your cue that the loop node's output won't wire cleanly into your custom sampler anymore.

Three modes, all driven by the internal per-mode counter:

• sequential - walks the list in order, wraps at the end.
• random - seeded choice using seed + step, so it's reproducible yet different every run.
• ping_pong - forward then backward, for neighbor-to-neighbor transitions.

reset (a BOOLEAN, default false) zeroes the counter when you want a fresh cycle.

The few inputs you'll actually set

• mode - sequential, random, ping_pong.
• skip_samplers - comma-separated names to exclude, e.g. lcm, deis. Exact, case-sensitive matches only; the console logs a warning for anything unrecognized, and if you skip everything it silently uses the full list.
• seed - matters only in random mode.

Outputs: sampler (SAMPLER type → your custom node's sampler input), current_index, total_combinations, and current_combination, the last being a "Sampler: euler" string that's nice for filenames or a text/PreviewAny label.

Install and the one gotcha

No dependencies, no model files - just ComfyUI's own sampler list under the hood. Install via ComfyUI Manager (search comfyui-ksampler-tester-loop) or:

cd ComfyUI/custom_nodes
git clone https://github.com/KY-2000/comfyui-ksampler-tester-loop

then restart. (You only ever install this once even if you use several nodes from the pack - they all ship together.)

The gotcha, again, is the name: "Loop" here means one selection per queue run. Queue the workflow repeatedly and each run hands you the next sampler object. It will not iterate internally and render a grid in a single execution - that's the pack's one consistent footgun, and the README is admirably clear about it ("each execution will use the next combination").

If you're testing for a stock KSampler, grab the non-Advanced Sampler Loop instead - no point paying the object-resolution overhead for a string input. Reach for this variant only when your sampler node actually needs the SAMPLER type.