ComfyUI Node

Ultimate Face Fix

Detects, reconstructs, semantically masks, and seamlessly blends one or many faces.

By Merserk·Created about a month ago·Updated about a month ago· 19
Ultimate Face Fix
  • image
  • model
  • vae
  • positive
  • negative
  • face_detector
  • face_landmarker
  • bboxes
  • face_parser
  • sam_model
  • fixed_image
  • original_face_crops
  • processed_face_crops
  • face_mask
  • debug_preview
face_selectionall
detection_qualitymaximum
max_faces16
min_face_size24
detection_confidence0.25
context_scale1.80
target_resolution1024
repair_modecustom
custom_denoise0.40
seed0
steps25
cfg5.5
sampler_nameeuler
schedulerbeta
mask_presetcore_face
sam_refineauto
mask_grow_percent1.5
mask_feather_percent2.5
color_match_strength0.65
blend_modemultiband
Categoryultimate face fix

Inputs (30)

NameTypeDefaultDescription
imageIMAGE
modelMODELGeneration model used for crop img2img.
vaeVAE
positiveCONDITIONING
negativeCONDITIONING
face_selectionCOMBOall3 options: all, largest, largest_n
detection_qualityCOMBOmaximum2 options: maximum, balanced
max_facesINT161–64
min_face_sizeINT244–1024
detection_confidenceFLOAT0.250.01–1
context_scaleFLOAT1.801.1–3
target_resolutionCOMBO10245 options: none, 512, 768, 1024, 1536
repair_modeCOMBOcustom4 options: detail, repair, reconstruct, custom
custom_denoiseFLOAT0.400–1
seedINT00–18446744073709550000
stepsINT251–200
cfgFLOAT5.50–100
sampler_nameCOMBOeuler44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38
schedulerCOMBObeta9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3
mask_presetCOMBOcore_face3 options: core_face, portrait_safe, head
sam_refineCOMBOauto3 options: auto, always, off
mask_grow_percentFLOAT1.50–15
mask_feather_percentFLOAT2.50–15
color_match_strengthFLOAT0.650–1
blend_modeCOMBOmultiband2 options: multiband, alpha
face_detectoroptFACE_FIX_DETECTORConnect Load Face Fix Detector. It may be omitted only when bboxes is connected.
face_landmarkeroptFACE_DETECTION_MODELConnect ComfyUI's Load Face Detection Model (MediaPipe) output for landmark-aware crops.
bboxesoptBOUNDING_BOX[object Object]When connected, these boxes replace YOLO detection.
face_parseroptFACE_FIX_PARSERConnect Load Face Fix Parser. When omitted, SAM/landmark/geometry fallback masks are used.
sam_modeloptMODELConnect the MODEL output from the supplied SAM 3.1 checkpoint.

Outputs (5)

NameTypeDescription
fixed_imageIMAGE
original_face_cropsIMAGE
processed_face_cropsIMAGE
face_maskMASK
debug_previewIMAGE