Stack_Ksampler_basic

This is Apt_Preset's take on ComfyUI's stock KSampler, stripped to the four settings most generations actually need: steps, CFG, sampler, and scheduler. No model, no positive/negative, no latent - those get wired in elsewhere by the pack's own consuming node. What you're configuring here is purely "how the denoising loop behaves," packaged as one KS_STACK output instead of four loose widgets on a bigger node.

The four fields

• steps (default 8, range 0–10000) - how many denoising steps to run.
• cfg (default 8, range 0–100) - classifier-free guidance strength.
• sampler (default euler, 44 choices) - the algorithm deciding how noise gets removed each step. The full list is essentially ComfyUI's stock roster: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, dpm_2, dpm_2_ancestral, lms, dpm_fast, and roughly 32 more the node menu doesn't list in full - the usual DPM++, UniPC, and DEIS variants included. Nothing exotic like RES4LYF's custom samplers; for those you'd want Stack_Ksampler_custom instead.
• scheduler (default normal, 9 choices: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, normal, linear_quadratic, kl_optimal) - the noise schedule shape.

Output: ksample, a KS_STACK object for the pack's own sampler-consuming node.

Read this before trusting the defaults

steps = 8 and cfg = 8 together don't match any single documented recipe cleanly - 8 steps is low for a non-distilled SD 1.5/SDXL model (those typically want 20–30) but the CFG-8 default doesn't match a distilled/turbo model's needs either (those generally want CFG 1, guidance already baked in). Whatever base model you're actually running, check its own recommended steps/CFG rather than trusting this node's out-of-the-box numbers. As a rule of thumb: standard SD 1.5 or SDXL checkpoints want 20–30 steps at CFG 5–9; Flux Dev wants 20–30 steps with its own guidance value rather than CFG; distilled/turbo models (Z-Image Turbo, Flux 2 Klein, ERNIE Turbo) want their few native steps (4–8) at CFG 1, and cranking CFG up on those doesn't add prompt adherence - it applies the correction twice and can burn the image.

Same caution applies to scheduler: karras is the safe default for SD 1.5/SDXL, and actively the wrong choice on flow-matching models like Flux, Z-Image, or Klein, where it over-concentrates denoising effort on a trajectory that's already close to a straight line. On those, normal, simple, or beta tend to behave better - check the model card first.

Basic vs the rest of the family

Reach for this node for the common case: one model, one pass, nothing fancy. Stack_Ksampler_adv adds partial-step sampling and noise-leftover controls for two-stage pipelines; Stack_Ksampler_custom swaps the dropdowns for wired-in NOISE/GUIDER/SAMPLER/SIGMAS objects if you're using third-party samplers (RES4LYF and similar).

Install

ComfyUI Manager: search "ComfyUI-Apt_Preset," install, restart. Manually: cd ComfyUI/custom_nodes && git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git, then restart. The install.bat dependency script is Windows-only - on Linux or a cloud box, read what it lists and pip-install those packages, or let ComfyUI Manager's "install missing nodes" flow handle it.

Troubleshooting

Underbaked, mushy-looking output is usually just too few steps for your model - bump steps up before touching anything else.

This pack has real, reported import failures - a comfyui subreddit thread shows (IMPORT FAILED): ...ComfyUI-Apt_Preset in someone's console. ComfyUI Manager's "install missing custom nodes" fixed it there, as did manually re-cloning the repo and restarting.