PPMSamplerSelect

PPMSamplerSelect is the pack's general-purpose sampler builder - it outputs a SAMPLER for SamplerCustom, with a couple of ppm-specific toggles baked in, most notably an optional CFG++ mode. Think of it as pamparamm's answer to the stock KSamplerSelect, but with the pack's extra sampler tricks available from one node.

What makes it different from the stock sampler select

Two things stand out in the inputs, and they're the reason you'd pick this over ComfyUI's built-in sampler node.

First, it takes a model input, which a plain sampler-select doesn't. That's there because some of the behaviors it can enable are model-aware - the sampler needs to know something about the model to do its thing correctly, rather than being a purely generic step function.

Second, there's a cfg_pp toggle for CFG++ mode. CFG++ is a reformulation of classifier-free guidance that greatly reduces the overexposure/oversaturation you get from normal CFG at higher scales. Flip it on and you're meant to run a low CFG (roughly 1-2) instead of the usual 7-8 - same guidance effect, far less blown-out contrast. It's the same family as the dedicated CFGPPSamplerSelect in this pack, folded in here as a switch.

The inputs

• sampler_name (dropdown) - pick the base sampler.
• model (MODEL) - wire your model in; the node uses it for model-aware sampling.
• cfg_pp (boolean, default false) - off is normal sampling; on enables CFG++ mode. If you turn this on, drop your CFG scale down to ~1-2 at the sampler, or you'll overcorrect.
• s_sigma_diff (default 2) - a fine-tuning knob for the dynamic step behavior. Leave it at the default unless you're deliberately experimenting.

Output is a SAMPLER → into SamplerCustom (this doesn't plug into a plain KSampler).

Install

No models, no deps:

• ComfyUI Manager: search "ComfyUI-ppm", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/pamparamm/ComfyUI-ppm, then restart.

Honest take

Samplers are the fashion district of this hobby and it's easy to overthink them, so here's the practical read. The headline feature is CFG++, and it's a legitimately good tool if your SDXL renders come out overcooked - the reduced overexposure is real, not placebo. The catch that bites everyone the first time: CFG++ needs a low CFG. People enable cfg_pp, leave CFG at 8, and get a washed-out or weird result, then blame the node. Set CFG to 1.0-2.0 when cfg_pp is on and it clicks.

If you don't need CFG++, honestly the stock sampler nodes are fine and there's no reason to reach for this - it earns its slot when you specifically want the pack's guidance behavior from a single node. And because it needs SamplerCustom downstream and a model input, it's a touch more wiring than the basic path; worth it only if you're using what it offers.

Note this node is thinly documented in the README compared to the pack's headline nodes - treat the CFG++ behavior as the reason it exists and A/B everything against your normal setup with a fixed seed. Pack rule: erroring after an update? Delete and re-add.