⚜️ KSampler Settings

This one doesn't get a shout-out in the FOCUS Nodes README - the author spotlights the All-In-One, the Control Module, the Model Unloader and a handful of others by name, and this quiet utility isn't among them. But it's built on the exact same instinct that drives the rest of the pack: stop scattering the same three dropdowns across five different nodes and just put them in one place.

What it actually does

It doesn't load a checkpoint or run a sampling step. It's a picker - you set a checkpoint, a sampler algorithm, and a scheduler curve once, and it hands those three choices out as outputs for the rest of your graph to consume. Think of it as a settings object you can duplicate rather than a functional node.

That's genuinely handy in this pack's signature move: a two-stage SDXL workflow where the creator swaps checkpoints mid-generation (their own example runs AlbedoBase XL for composition, then a handful of refinement steps on Dreamshaper). Declaring your sampler/scheduler combo once here and copying the node, instead of re-setting three widgets from scratch on a second KSamplerAdvanced, is exactly the kind of screen-clutter reduction the pack is going for.

Inputs and outputs that matter

• ckpt_name - an enum pulled straight from your models/checkpoints folder, same list ComfyUI's own checkpoint loader shows. There's nothing pack-specific baked in; it just mirrors what's on disk.
• sampler_name - the full 34-entry sampler roster ComfyUI ships: euler, the dpm_2/dpmpp family, heun, ddim, uni_pc, the res_multistep variants and their cfg_pp cousins. If you're not sure which one to reach for, DPM++ 2M Karras is still the safe default for SD 1.5 and SDXL - this pack's whole world - and Euler a is the go-to for the anime-leaning SDXL finetunes.
• scheduler_name - 9 choices: normal, karras, exponential, sgm_uniform, simple, ddim_uniform, beta, linear_quadratic, kl_optimal. Karras is the one glued onto most SD 1.5/SDXL sampler names for a reason - it concentrates denoising effort where it matters at lower step counts.

Outputs mirror the inputs: Checkpoint (the filename, as a plain value you can route to a loader or note node), Sampler and Scheduler (typed to match ComfyUI's own combo widgets, so they plug straight into a KSampler once you've converted its sampler_name/scheduler widgets to inputs).

Installing it

Via ComfyUI Manager: search Comfyui FOCUS nodes (the exact repo name works more reliably than a loose "focus" search - a couple of early users reported the manager not finding it by title alone). Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/DJ-Tribefull/Comfyui_FOCUS_nodes

Restart ComfyUI. No models to download for this node specifically - it just reads what you've already got installed.

Common issues

ckpt_name dropdown is empty. That means ComfyUI can't see anything in models/checkpoints. Drop a .safetensors file in there and restart - this node has no fallback list of its own.

Wiring the outputs into a stock KSampler does nothing. By default, sampler_name and scheduler are widgets on the KSampler node, not input sockets. Right-click the KSampler, use "Convert Widget to Input" on both, and then the noodles from this node's Sampler/Scheduler outputs will actually connect.

You only need this if you're duplicating settings across stages. For a single-checkpoint, single-pass workflow, a plain checkpoint loader and the widgets already on your KSampler do the same job with one fewer node in the graph. This one earns its place specifically once you've got more than one place that needs to agree on the same sampler/scheduler/checkpoint choice.