Smart Sampler Settings

Sampler settings are the most-tweaked, most-shared values in ComfyUI, and most workflows scatter them across a dozen KSampler widgets. Smart Sampler Settings centralizes them: one node holds sampler, scheduler, steps, CFG, denoise, guidance, noise injection, an upscale factor, and a seed - and emits the enabled subset as a single pipe that any Eclipse-aware sampler or context node can consume. Change it in one place, and every consumer in the graph sees it.

The clever bit is selective output. Every parameter group is a combo-chip: enabled chips get included in the pipe, disabled ones are excluded entirely. So a "steps + cfg only" pipe is genuinely different from a "everything" pipe, and downstream nodes only ever receive values you explicitly configured - no phantom defaults leaking in. It replaces a whole family of legacy "Sampler Settings" nodes with one interface.

How it works

The features input (comma-separated; the frontend turns it into chips) drives everything: sampler, scheduler, steps, cfg, denoise, guidance, noise_injection, upscale, seed, plus allow_overwrite. Only enabled features make it into the pipe's dictionary.

The settings that matter:

• sampler_name / scheduler - the algorithm picks. These are the two you'll change most; on newer models the sampler/scheduler pairing matters a lot more than CFG fiddling.
• steps, cfg, denoise - the classics. CFG on modern distilled models behaves differently than on SDXL-lineage ones, which is exactly why having it centralized (and easy to sweep across a workflow) is handy.
• guidance - a separate guidance value for models that use it (Flux-style).
• sigmas_denoise / noise_strength - the noise-injection pair for advanced sampling (the noise_injection chip).
• upscale_value - a scale factor for upscale workflows.
• seed - with the usual -1 random / -2 increment modes.
• allow_overwrite - when on, lets direct inputs to downstream IO nodes overwrite this node's values. Off means this node's values win. This is the flag people either never notice or desperately need.

The output is the pipe (a dict) with just the selected keys.

Where it fits

It's the configuration heart of an Eclipse pipe workflow: wire the pipe into a context node, a sampler, or Pipe IO Sampler Settings for inspection. If you're on the Eclipse pipe ecosystem, this is the canonical place to define generation settings once. If you're not on pipes, it's still the cleanest way to make sampler settings a single, visible, shareable block.

Installing it

Part of ComfyUI_Eclipse:

git clone https://github.com/r-vage/ComfyUI_Eclipse custom_nodes/ComfyUI_Eclipse

Or ComfyUI Manager → search "ComfyUI_Eclipse" → install and restart. No models.

Gotchas

The pipe only contains what you enabled - that's a feature, but it means a downstream node that expects a value you forgot to chip in will fall back to its own default instead of yours, which can look like "this setting isn't working." Check the chips before assuming a bug. And allow_overwrite is easy to misread: it's the other nodes' permission, not this node's. Old RvTools-era workflows predating v4.0.0 need the pack's migration tool; new ones don't.