KSamplerAdvancedProvider

This node doesn't sample anything itself - it's a "provider," Impact Pack's pattern for bundling a sampler configuration into a single typed object that a more complex node consumes, rather than repeating a dozen widgets on every node that needs sampling settings. The README says it plainly: it's a wrapper that lets a KSamplerAdvanced-style configuration be used inside TwoAdvancedSamplersForMask and RegionalSampler/RegionalSamplerAdvanced.

Where this actually matters

Skip this unless you're building something specific: a workflow where different regions of an image need different sampling - a masked area sampled with one config, the rest of the image with another, or several regions each getting their own sampler within a single pass. That's what RegionalSampler and TwoAdvancedSamplersForMask are for, and both need their per-region sampler settings handed to them as a KSAMPLER_ADVANCED object rather than as loose widgets. This node builds that object.

If you're running a normal single-pass generation, or even a normal FaceDetailer pass, you'll never touch this - it's specifically for the pack's regional-sampling machinery.

The inputs and output

• cfg (default 8, range 0–100), sampler_name (44 choices - the same catalog a regular KSampler offers, from euler through the CFG++ variants and beyond), scheduler (17 choices, including the newer AYS SDXL/SD1/SVD variants, kl_optimal, and linear_quadratic) - the standard sampling trio.
• sigma_factor (default 1, range 0–10, step 0.01) - multiplies the denoise schedule. Per the README, this is how you scale how aggressively this particular sampler denoises relative to whatever denoise value the consuming node applies - useful when you want a masked region sampled at a different effective strength than the base image in the same pass.
• basic_pipe (BASIC_PIPE) - the bundled model/clip/vae/positive/negative package. You need this built already, typically via ToBasicPipe, before this node can do anything.
• Optional: sampler_opt (SAMPLER) to override the built-in sampler catalog with a custom one, scheduler_func_opt (SCHEDULER_FUNC) to override the scheduler with a custom function (the pack's GITSScheduler Func Provider is one example of what plugs in here).

Output is a single KSAMPLER_ADVANCED - feed it directly into TwoAdvancedSamplersForMask's or RegionalSampler's corresponding sampler input.

Installing it

Bundled with the pack. Via ComfyUI Manager: search ComfyUI Impact Pack, Install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Impact-Pack
cd ComfyUI-Impact-Pack
pip install -r requirements.txt

(portable: python_embeded\python.exe -m pip install -r requirements.txt), restart. No extra models for this node itself - it just packages settings around whatever checkpoint your basic_pipe already carries.

Common issues

No BASIC_PIPE to plug in. This node wants a bundled pipe, not loose model/clip/vae connections - build one with ToBasicPipe off your checkpoint loader and text encodes first.

Wired into the wrong consumer. This produces KSAMPLER_ADVANCED specifically, which is a different type from the plain KSAMPLER that the non-advanced KSamplerProvider produces. RegionalSamplerAdvanced and TwoAdvancedSamplersForMask want the advanced type; RegionalSampler and TwoSamplersForMask want the plain one. Get them backwards and ComfyUI will refuse the connection outright at prompt time - a visible, catchable error rather than a silent misconfiguration, so at least you'll know immediately.

Results look over- or under-denoised in a regional setup. Check sigma_factor before assuming your cfg or sampler choice is wrong - it's the knob specifically meant to scale denoising strength for this sampler relative to the rest of the pipeline, and it's easy to forget it's there since a plain KSampler has no equivalent.