Earngenix Logo
Skip to main content

Glossary · ComfyUI Basics

What Is KSampler in ComfyUI?

By Earngenix Team ·

⚡ Quick Answer

KSampler is the ComfyUI node that actually generates your image. It takes your model, positive and negative conditioning, and a starting latent, then runs the denoising process for a set number of steps to produce the finished latent.

Every other node in a basic workflow exists to feed KSampler what it needs — the checkpoint, the prompt, the empty canvas. KSampler is where those pieces actually get combined and the noise-to-image process happens, controlled by widgets like steps, cfg, sampler_name, scheduler, and denoise.

Where You'll See It

Sitting in the middle of nearly every workflow, wired from an Empty Latent Image (or an image-to-image source) on one side, a Load Checkpoint and two CLIP Text Encode nodes for conditioning, and outputting to a VAE Decode node on the other side to turn the result into a viewable image.

Quick Example

A typical SDXL text-to-image KSampler runs at around 25–30 steps, cfg 6–8, sampler_name dpmpp_2m, scheduler karras, and denoise 1.0 for a full generation from an empty latent.

Use KSampler (Advanced) instead of the standard node when you need to split sampling across two connected KSamplers — for example, running the first steps on one model and finishing on another.

Common Mix-Up

KSampler vs. sampler: KSampler is the node itself. "Sampler" (set via the sampler_name widget, like euler or dpmpp_2m) is the specific mathematical method KSampler uses internally to remove noise at each step — the node is the container, the sampler is one setting inside it.

Frequently Asked Questions

KSampler runs the full denoising process in one pass with a single seed. KSampler (Advanced) exposes start_step and end_step separately, letting you split sampling across multiple KSamplers — useful for techniques like switching models partway through generation.

Steps set how many denoising passes KSampler runs. Too few steps means the sampler hasn’t had enough passes to fully resolve detail, which often shows up as blurry or muddy results, especially with slower samplers.

Yes, though less than steps and cfg for most everyday generations. Different samplers (Euler, DPM++ 2M, UniPC, and others) trade off speed, stability, and how much they change with the same seed — worth testing a few on a workflow you use often.

See It In Action

Ready to run your first generation?

Our text-to-image tutorial walks through every KSampler setting step by step.

Published: 2026-09-04 · Last updated: 2026-09-04

Discussion

Join the discussion

Sign in to leave a comment or reply

💬

No comments yet

Be the first to share your thoughts!