Flux Settings Node

The name undersells it. FluxSettingsNode isn't a settings storage node - it's four core ComfyUI nodes welded into one: FluxGuidance, KSamplerSelect, BasicScheduler, and RandomNoise. You feed it a model and a conditioning, and it hands back the CONDITIONING, SAMPLER, SIGMAS, and NOISE you'd normally assemble from four separate nodes. On its own that saves you maybe a minute of graph wiring. The actual selling point is the tabs: six of them sit on top of the node, each remembering its own sampler, scheduler, steps, guidance, and denoise. That's six full configs you can flip between by clicking a number.

If you A/B test prompts or sampler settings at all, you'll feel why that matters. Normally each comparison means dragging values around in a cramped KSampler widget, or building five near-identical sampler chains. Here you dial in preset 1, click 2, dial in a different batch of settings, and bounce between them mid-session. The presets persist into your saved workflow too, since the tab state is serialized with the node.

How it actually works

The mechanism is refreshingly boring, and that's a compliment. In the node's Python it does exactly what the four native nodes do:

• guidance gets written into the conditioning via conditioning_set_values(..., {"guidance": ...})
• sampler_name resolves through ComfyUI's SAMPLER_NAMES registry
• sigmas come from calculate_sigmas with the model's own model_sampling, so it always uses the schedule your checkpoint expects
• noise comes from RandomNoise seeded with noise_seed

Everything is standard library calls; there's no custom math to go wrong. The tab UI is frontend JavaScript that saves those five widget values on switch and restores them on click. One genuine gotcha: the tabs only remember guidance, sampler_name, scheduler, steps, and denoise - noise_seed is shared across all six. That's actually convenient for comparing settings on a fixed seed, but don't expect a per-tab seed.

Also note the "Flux" in the name. The guidance field is Flux-flavored - it injects the guidance value into conditioning, which only Flux-family models actually read. Drop this on an SDXL checkpoint and the guidance input quietly does nothing. The sampler and scheduler part still works on anything, so it's a passable generic custom-sampling shortcut, but it's built for Flux.

The inputs that matter

Three of them, really. The rest are pass-through or fine at defaults.

• guidance (default 3.5) - Flux Dev's canonical guidance, though plenty of people argue 3.5 runs a bit hot and nudge toward 2.5–3. Higher buys prompt adherence at the cost of a plastic look.
• scheduler - for Flux, the short answer is "not Karras." Flow-matching models want a conservative schedule: normal or beta are the usual picks, simple works too. Karras will actively fight the model.
• sampler_name - euler is the safe Flux default. You can use denoise under 1.0 for img2img; the node recomputes the schedule across more steps and trims to your requested count, same as BasicScheduler.

The four outputs wire straight into a model-sampling area of your graph: CONDITIONING and SIGMAS into a SamplerCustom or KSampler-style node, SAMPLER and NOISE into their matching inputs.

Installing it

There are no dependencies and no model files - the source only imports core ComfyUI modules, so this is one of the rare custom nodes that can't drag your environment into dependency hell. Either route works:

cd ComfyUI/custom_nodes
git clone https://github.com/Light-x02/ComfyUI-FluxSettingsNode

then restart ComfyUI. Or use ComfyUI Manager, search "FluxSettingsNode," install, restart. That's the whole install. The pack also ships a tiny DisableNoise node (empty noise instead of random) if you ever want deterministic-seed-free debugging.

Things to know before you lean on it

The author advertises compatibility with ComfyUI-ImageMetadataExtension so sampler/steps/guidance/prompt/seed land correctly in Civitai metadata. That's a real nicety for people who share workflows, but it only matters if you've installed that extension - the node does nothing special on its own.

It's a young, small pack (a 1.1.1, a Ko-fi link, MIT tab code borrowed from ComfyUI-Flux-Continuum), and the community signal around it is basically nil - I couldn't find any real user discussion of it. So treat it as a handy quality-of-life node, not a battle-tested essential. If you're rebuilding a Flux pipeline from scratch you don't need it; if you're tired of reaching into KSampler widgets forty times a session, it earns its keep. The tab pattern is a great idea - just know you're an early adopter.