Skip to content

KSampler Node ​

Sample latent image using model and prompts.

Function ​

Core sampling node that generates images from noise.

Inputs ​

InputTypeDescription
modelmodelModel from CheckpointLoader
positiveconditioningPositive prompt encoding
negativeconditioningNegative prompt encoding
latentlatentLatent image to sample

Outputs ​

OutputTypeDescription
LATENTlatentSampled latent image

Parameters ​

ParameterTypeDescription
seedintRandom seed
stepsintSampling steps (20-50)
cfgfloatCFG scale (7-10)
sampler_namedropdownSampler algorithm
schedulerdropdownScheduler type
positiveconditioningPositive conditioning
negativeconditioningNegative conditioning

Sampler Types ​

SamplerSpeedQualityNotes
eulerFastGoodRecommended
euler_ancestralFastGoodMore variation
dpm++ 2MMediumBetterSmooth
dpm++ 2M KarrasMediumBestConsistent
ddimSlowGoodTraditional
UniPCFastGoodNew algorithm

Schedulers ​

SchedulerEffect
normalStandard
karrasTighter at end
exponentialSmoother
simpleBasic

Parameters Explained ​

Steps ​

ValueUse Case
10-15Quick preview
20-30Standard quality
35-50High quality

CFG Scale ​

ValueEffect
3-5Creative, less strict
6-8Balanced
9-12Strict prompt following

Usage Example ​

CheckpointLoader
    ↓ MODEL → KSampler
    ↓ CLIP
CLIPTextEncode (positive) → KSampler
    ↓
CLIPTextEncode (negative) → KSampler
    ↓
EmptyLatentImage → KSampler
    ↓ LATENT
VAEDecode → SaveImage

Tips ​

  1. Start with defaults: 20 steps, CFG 8
  2. Increase for detail: More steps = more detail
  3. Adjust CFG: Higher for strict following

Troubleshooting ​

IssueSolution
Black imageCheck model/VAE
Garbled outputLower CFG
Too similarIncrease steps

Next Steps ​