Shima Sampler Commons Passer

The Shima Sampler Commons Passer does one thing: it takes the shima.samplercommons bundle from a Shima Sampler Commons node and re-emits it, plus the five individual settings it contains (steps, cfg, sampler, scheduler, denoise) on separate sockets. It's the plumbing equivalent of an extension cord - no logic, no defaults, just the bundle arriving somewhere else, ready to be unpacked.

If you've ever wanted to read the settings a Sampler Commons computed and still pass the whole bundle along to a Shima Sampler in another part of the graph, this is the node that lets you do both without splitting a wire.

How it works

One required input - the shima.samplercommons DICT bundle - and six outputs:

• shima.samplercommons - the bundle, unchanged.
• steps (INT), cfg (FLOAT), sampler_name (STRING), scheduler (STRING), denoise (FLOAT) - the unpacked values.

Because the bundle is a plain dict, the pass is trivial: take the dict, hand it back, and also look up each key for the individual outputs. The individual outputs are what make the node useful - they let a plain KSampler or a note node consume the settings even if it knows nothing about Shima bundles. It's the same "airport" pattern as the pack's other passers: an optional-input-friendly, type-stable landing pad for Use Everywhere broadcasts, plus a way to fan a bundle's contents out to standard nodes.

The inputs that matter

Just the one: shima.samplercommons. If you connect a Shima Sampler Commons bundle, everything downstream lights up. The allow_external_linking toggle (off by default) controls whether the node broadcasts outside its island group.

Where it fits

Two real uses. First, fan-out: Sampler Commons computes one canonical set of settings, and you want both a Shima Sampler (which consumes the bundle) and a non-Shima sampler or a display node (which needs the individual values). Second, readability: when a long bundle wire crosses the canvas, a passer at the far end gives you a clean socket to aim at and a way to see the values mid-graph without squinting at the source.

Install

Same as the whole pack:

cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf.git Shima
pip install -r requirements.txt

Restart, then Shima/Utilities/Passers → Shima Sampler Commons Passer. ComfyUI Manager: search "Shima". First boot auto-installs ComfyUI-Impact-Pack and cg-use-everywhere if absent - the Use Everywhere dependency, expected.

Common issues

The only real failure mode is feeding it something that isn't a shima.samplercommons dict - any DICT will type-check, but the individual outputs will come back as None if the keys don't exist. It passes through what it's given, including mistakes: a stale bundle (settings computed before you changed the model type) will still be stale after the pass. There's nothing here to fix beyond wiring it to the right source.