👑 FRED Simplified Parameters Panel

Most sampling parameters in ComfyUI are spread across separate widget rows on the KSampler. FRED_Simplified_Parameters_Panel pulls the core ones - scale, denoise, guidance, steps, seed, sampler, scheduler - into one clean panel, and then goes further than a glorified slider group: it actually constructs the internal objects a sampler consumes. It outputs a proper NOISE object, a SAMPLER object, and a SIGMAS tensor, pre-computed for your model. Wire them into a sampler node and you're using advanced sampling internals without touching a single math node.

The inputs

• model - the panel needs it to compute sigmas for your model's specific scheduler.
• scale (1.4), denoise (0.6), guidance (2.2), steps (8), noise_seed (42) - the familiar knobs.
• sampler - 44 sampler choices (euler, euler_ancestral, dpm_2, and the whole extended family).
• scheduler - 9 choices including simple, sgm_uniform, karras, exponential.
• custom_widget_names (optional) - a string used to hint the model name detection if your model wire doesn't resolve cleanly.

The outputs

• scale, denoise, guidance, steps, seed - pass-throughs as plain values, so you can drive other nodes from the same panel.
• noise (NOISE) - a proper Noise_RandomNoise object seeded from noise_seed. This is the real deal, not a float: you can feed it into a KSampler's noise input.
• sampler (SAMPLER) and sampler_name (STRING) - the sampler object plus its name for metadata.
• sigmas (SIGMAS) - the actual sigma schedule tensor for the chosen scheduler, computed for your model with total_steps correctly adjusted when denoise < 1.0. Drop it into a sampler's sigmas input and you have full control.
• scheduler_name (STRING) - for the metadata record.
• model_name (STRING) - the panel sniffs your connected model and reports which checkpoint it came from, which is a nice freebie for the FRED Image Saver's %model_name token.

That combination - real NOISE, SAMPLER, and SIGMAS objects plus the plain values - means the panel can drive either a stock KSampler or an advanced sampler node, and it doubles as a tidy way to hand sampler objects to nodes that want them.

Why you'd use it

Two audiences. Quick prototyping: instead of hunting through a KSampler's collapsed widget rows, one panel holds the few things you actually tune, and the pass-through outputs let you mirror them elsewhere. Advanced setups: it's a legitimate way to get a scheduler-correct SIGMAS tensor and a real NOISE object for custom sampling graphs without reimplementing calculate_sigmas yourself. The model-name sniffing also quietly solves a metadata problem - you don't have to hand-type the checkpoint name into your saver.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Poukpalaova/ComfyUI-FRED-Nodes_v2.git

Restart, or "ComfyUI FRED Nodes v2" via ComfyUI Manager. Requires torch (bundled with ComfyUI) and the pack's base dependencies.

Gotchas

The defaults (scale 1.4, guidance 2.2, steps 8) look Flux-oriented, so check them if you're on SDXL or SD1.5 - steps 8 with denoise 0.6 will give you a very fast, very loose result on models that want 20+. And the model-name sniffing is best-effort: if your loader is wrapped in an unusual custom node, model_name may come back empty and you'll want custom_widget_names or a manual override.