Nodes/CRT-Nodes/Ultralytics Enhancer (CRT)
ComfyUI Node

Ultralytics Enhancer (CRT)

A ComfyUI node in CRT/Sampling with 34 inputs and 5 outputs.

By PGCRT·Created 2 years ago·Updated 7 days ago· 129
Ultralytics Enhancer (CRT)
  • image
  • model
  • positive
  • negative
  • vae
  • enhanced_image
  • enhanced_face
  • cropped_face_before
  • enhanced_face_alpha
  • base_face_alpha
face_segm_model
segm_threshold0.50
grow_crop0
sampler_namelcm
schedulersimple
cfg1.0
steps2
denoise1.00
seed1
edit_model_flux2kleinfalse
upscale_megapixel1.5
resize_back_to_originalfalse
multi_face_resolution_strategyoptimal
pre_crop_factor1.00
post_mask_expand10
post_mask_blur12.0
post_mask_taper_borders8
stage1_sigma_factor1.010
stage2_sigma_factor0.990
stage1_sigma_start0.100
stage2_sigma_end1.000
details_amount_stage10.05
details_amount_stage20.10
enable_noise_injectionenable
injection_point0.50
injection_strength0.05
normalize_injected_noiseenable
enhancement_mix1.00
color_match_strength1.00
CategoryCRT/Sampling

Inputs (34)

NameTypeDefaultDescription
imageIMAGE
face_segm_modelCOMBO0 options:
segm_thresholdFLOAT0.500–1
grow_cropINT00–512Expands each detected crop region by this many pixels before enhancement
modelMODEL
positiveCONDITIONING
sampler_nameCOMBOlcm44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38
schedulerCOMBOsimple9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3
cfgFLOAT1.00–100Classifier Free Guidance scale. Higher values follow the prompt more closely.
stepsINT21–100
denoiseFLOAT1.000–1Amount of denoising to apply. 1.0 = full denoising (txt2img), 0.5-0.8 typical for img2img.
seedINT10–18446744073709550000
edit_model_flux2kleinBOOLEANfalseEnable per-crop reference latent conditioning for edit models like flux2klein
upscale_megapixelFLOAT1.50.1–16Target megapixels used to upscale each crop before enhancement
resize_back_to_originalBOOLEANfalseIf disabled, the whole input is upscaled so refined crops can keep higher detail
multi_face_resolution_strategyCOMBOoptimalUsed only when resize_back_to_original is False to decide global upscale for multi-face inputs
pre_crop_factorFLOAT1.000.1–1Tightens each SEGS crop around its center before enhancement. Lower is tighter.
post_mask_expandINT10-64–64
post_mask_blurFLOAT12.00–64
post_mask_taper_bordersINT80–128Fades the mask edges to black to prevent hard clipping lines
stage1_sigma_factorFLOAT1.0100–100
stage2_sigma_factorFLOAT0.9900–100
stage1_sigma_startFLOAT0.1000–1
stage2_sigma_endFLOAT1.0000–1
details_amount_stage1FLOAT0.05-5–5
details_amount_stage2FLOAT0.10-5–5
enable_noise_injectionCOMBOenableEnable noise injection during sampling
injection_pointFLOAT0.500–1Percentage of steps after which to inject noise
injection_strengthFLOAT0.05-20–20Strength of injected noise
normalize_injected_noiseCOMBOenableNormalize injected noise to match latent statistics
enhancement_mixFLOAT1.000–1Mix between original (0.0) and enhanced (1.0) face. 0.5 = 50/50 blend
color_match_strengthFLOAT1.000–2Strength of color matching between original and enhanced face. 0.0 = disabled, 1.0 = full matching
negativeoptCONDITIONING
vaeoptVAE

Outputs (5)

NameTypeDescription
enhanced_imageIMAGE
enhanced_faceIMAGE
cropped_face_beforeIMAGE
enhanced_face_alphaIMAGE
base_face_alphaIMAGE