KSamplerProvider

This node is a wrapper, and understanding why it exists is most of understanding it. Some Impact Pack nodes - TwoSamplersForMask, the upscaler providers - don't run sampling themselves; they orchestrate sampling and need you to hand them a sampler as an ingredient. But a normal ComfyUI KSampler is an all-in-one node that runs immediately; you can't pass "a KSampler" around as an object. KSamplerProvider fixes that. It bundles all the KSampler settings into a reusable KSAMPLER you can plug into those orchestrator nodes. The README puts it simply: "a wrapper that enables KSampler to be used in TwoSamplersForMask."

So you reach for it whenever a TwoSamplersForMask setup asks for a sampler and you want that sampler to be a plain KSampler with your chosen steps, cfg, and scheduler.

How it works

Instead of sampling, it captures a configuration. You set the seed, steps, cfg, sampler, scheduler, and denoise, feed it a basic_pipe (which carries the model and conditioning), and it emits a KSAMPLER object - a sampler-on-a-stick. The orchestrator node then invokes that sampler on whichever region or pass it's managing. Nothing generates here; you're defining a sampler for something else to use.

The inputs that matter

• basic_pipe - the model/clip/vae/positive/negative bundle the sampler will run with. Required, and the reason this node needs a pipe rather than loose model wires.
• steps (default 20), cfg (default 8), denoise (default 1) - the usual sampling dials. For a mask region that's blending into an existing base, you'll often lower denoise well below 1.
• sampler_name and scheduler - the full ComfyUI menus (euler, dpmpp variants, and the rest; simple/karras/exponential and Impact's added AYS schedulers).
• seed - sampling seed.

There's one optional input, scheduler_func_opt, for supplying a custom scheduler function (the pack's GITS/AYS tooling). Output is a single KSAMPLER - wire it into TwoSamplersForMask's base or mask sampler slot, not into a latent.

How to install it

ComfyUI Manager: search ComfyUI Impact Pack, install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/ltdrdata/ComfyUI-Impact-Pack, install requirements into ComfyUI's Python (pip install -r requirements.txt, or ..\..\..\python_embeded\python.exe -m pip install -r requirements.txt on Windows portable), restart. Auto-install ended at v7.6. A small SAM model downloads to ComfyUI/models/sams on first run.

Common issues

You wired it expecting an image or latent and got nothing. This node outputs a KSAMPLER, not a sampled result. Its whole job is to be consumed by another node (TwoSamplersForMask and friends). If you just want to sample a latent, use a regular KSampler - this wrapper is only for the mask/upscaler orchestrators.

Scheduler-to-input compatibility errors. If you converted the scheduler widget to an input elsewhere and it won't connect, the README points at the Impact Scheduler Adapter node to bridge that - a known wrinkle from the AYS schedulers being added to the menu.

Need the advanced sampler instead. For RegionalSampler and TwoAdvancedSamplersForMask, the counterpart is KSamplerAdvancedProvider, which exposes start/end step control and a sigma_factor. This basic provider pairs with the non-advanced TwoSamplersForMask.