technical

What are samplers and which should I use?

Samplers are algorithms that determine how the AI builds your image during generation. Different samplers produce different results and speeds.

Recommended samplers:

SamplerSpeedQualityBest For
DPM++ 2M KarrasFastExcellentGeneral use
Euler aFastGoodCreative variety
DPM++ SDE KarrasMediumExcellentDetailed images
DDIMFastGoodConsistent results
UniPCVery FastGoodSpeed priority

Quick recommendations:

  • Default choice: DPM++ 2M Karras (20-30 steps)
  • Need speed: UniPC or Euler (15-20 steps)
  • Maximum quality: DPM++ SDE Karras (25-40 steps)
  • Consistent outputs: DDIM (30-50 steps)

What affects your choice:

  • Available time (some samplers are 2-3x slower)
  • Whether you need reproducibility
  • Style of image you’re generating
  • Personal preference (test and compare)

Steps and samplers: Some samplers converge faster than others. DPM++ 2M Karras often looks great at 20 steps; DDIM may need 40+.

Start with DPM++ 2M Karras at 25 steps—it’s the most reliable default for most workflows.