Y
Prompt Workshop
Article · SD parameters

Stable Diffusion Sampler Guide

Samplers decide the algorithmic path from noise to image. Different samplers vary in convergence, detail and style tendency. This article focuses on the four that matter in production.

Four production samplers

SamplerBehaviorRecommended stepsBest for
Euler aFast, stylized, non-converging20–30Illustration, anime
DPM++ 2M KarrasDetail-rich, converging25–35Realistic portrait, product
UniPCFewest steps to acceptable10–20Quick drafting
DDIMDeterministic, reproducible30–50Reproducibility-critical work

Converging vs non-converging

"Converging" samplers (DPM++ 2M) stabilize and approach a fixed solution as steps increase. "Non-converging" samplers (Euler a, Ancestral) keep introducing noise — every extra step keeps changing the image. Converging samplers are easier to control in production.

Quick picks

  • Realistic portrait: DPM++ 2M Karras, 30 steps
  • Anime / illustration: Euler a, 25 steps
  • Quick draft: UniPC, 15 steps
  • Reproducibility: DDIM, 30 steps

Frequently asked questions

What does Karras mean?

Karras refers to a noise-schedule strategy. Most samplers ship a Karras variant (DPM++ 2M Karras, Euler Karras) — usually with better detail.

Which sampler does Flux use?

Flux typically uses Euler or default samplers; CFG is replaced by guidance.

What's the most stable choice on SDXL?

DPM++ 2M Karras is the safest default.

Try this parameter live in the editor

The local editor includes Midjourney / SDXL / Flux templates with a visible, editable parameter panel.

Open the editor →
Yan · AI Prompt Workshop editorial team|Last updated on 2026-06-12。This site does not call any cloud model. Every prompt and parameter in this article was tested and refined locally by the editorial team.