ComfyUI Node
DMC Face Detailer · Ordered Characters
A ComfyUI node in donut/create with 67 inputs and 6 outputs.
DMC Face Detailer · Ordered Characters
- image
- model
- clip
- vae
- positive
- negative
- bbox_detector
- sam_model_opt
- segm_detector_opt
- detailer_hook
- scheduler_func_opt
- edit_model
- face_reference
- face_reference_b
- nag_negative
- nag_ref_boost_mask
- image
- cropped_refined
- cropped_enhanced_alpha
- mask
- detailer_pipe
- cnet_images
◄resolution1024►
◄max_resolution0►
◄guide_size_fortrue►
◄seed0►
◄steps20►
◄cfg8.00►
◄sampler_name▾►
◄scheduler▾►
◄denoise0.50►
◄feather5►
◄noise_masktrue►
◄force_inpainttrue►
◄bbox_threshold0.50►
◄bbox_dilation10►
◄bbox_crop_factor3.0►
◄sam_detection_hint▾►
◄sam_dilation0►
◄sam_threshold0.93►
◄sam_bbox_expansion0►
◄sam_mask_hint_threshold0.70►
◄sam_mask_hint_use_negative▾►
◄drop_size10►
◄wildcard—►
◄cycle1►
◄max_faces2►
◄inpaint_modelfalse►
◄noise_mask_feather20►
◄edit_modefalse►
◄edit_prompt—►
◄edit_negative_prompt—►
◄grounding_px768►
◄vary_seed_per_facefalse►
◄turbo_modefalse►
◄nag_enabledfalse►
◄nag_phi4.0►
◄nag_tau2.50►
◄nag_alpha0.25►
◄nag_sigma_start1000.0►
◄nag_sigma_end0.0►
◄nag_ref_boost1.00►
◄nag_ref_boost_a1.00►
◄nag_fit_modefit►
◄nag_auto_phifalse►
◄nag_phi_scale1.00►
◄vae_damage_correctionfalse►
◄vae_damage_strength1.00►
◄nag_alpha_scheduleconstant►
◄nag_alpha_start0.25►
◄nag_alpha_end0.25►
◄face_prompts_json[]►
◄require_single_facefalse►
Categorydonut/create
Inputs (67)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| model | MODEL | — | |
| clip | CLIP | — | |
| vae | VAE | — | |
| resolution | INT | 1024256–4096 | Equivalent square target size. Sampling canvases are snapped to 64-pixel multiples. |
| max_resolution | INT | 00–4096 | Maximum edge length (0 = no limit). |
| guide_size_for | BOOLEAN | true | — |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 201–10000 | — |
| cfg | FLOAT | 8.000–100 | — |
| sampler_name | COMBO | 44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38 | |
| scheduler | COMBO | 9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3 | |
| positive | CONDITIONING | — | |
| negative | CONDITIONING | — | |
| denoise | FLOAT | 0.500.0001–1 | — |
| feather | INT | 50–100 | — |
| noise_mask | BOOLEAN | true | — |
| force_inpaint | BOOLEAN | true | — |
| bbox_threshold | FLOAT | 0.500–1 | — |
| bbox_dilation | INT | 10-512–512 | — |
| bbox_crop_factor | FLOAT | 3.01–10 | — |
| sam_detection_hint | COMBO | 9 options: center-1, horizontal-2, vertical-2, rect-4, diamond-4, mask-area, +3 | |
| sam_dilation | INT | 0-512–512 | — |
| sam_threshold | FLOAT | 0.930–1 | — |
| sam_bbox_expansion | INT | 00–1000 | — |
| sam_mask_hint_threshold | FLOAT | 0.700–1 | — |
| sam_mask_hint_use_negative | COMBO | 3 options: False, Small, Outter | |
| drop_size | INT | 101–16384 | — |
| bbox_detector | BBOX_DETECTOR | — | |
| wildcard | STRING | — | |
| cycle | INT | 11–10 | — |
| max_faces | INT | 21–20 | Maximum number of faces to process (largest by area) |
| sam_model_optopt | SAM_MODEL | — | |
| segm_detector_optopt | SEGM_DETECTOR | — | |
| detailer_hookopt | DETAILER_HOOK | — | |
| inpaint_modelopt | BOOLEAN | false | — |
| noise_mask_featheropt | INT | 200–100 | — |
| scheduler_func_optopt | SCHEDULER_FUNC | — | |
| edit_modeopt | BOOLEAN | false | Use Krea2 identity edit conditioning for each detected face. |
| edit_promptopt | STRING | — | |
| edit_modelopt | MODEL | Optional Krea2 model with the Identity Edit LoRA already applied. Falls back to model. | |
| face_referenceopt | IMAGE | Required in edit mode. The bbox detector extracts the identity face from this image. | |
| edit_negative_promptopt | STRING | — | |
| grounding_pxopt | INT | 7680–4096 | — |
| vary_seed_per_faceopt | BOOLEAN | false | Use a unique seed offset for each detected face. |
| turbo_modeopt | BOOLEAN | false | Snap denoise to a valid Turbo scheduler point. |
| face_reference_bopt | IMAGE | Optional subject/identity image for two-reference edits. Faces are extracted from this image when possible; if it has no detectable face, the detailer falls back to face_reference. | |
| nag_enabledopt | BOOLEAN | false | Apply Krea2 NAG inside sampling (requires krea2-nag). Uses CFG 1; Turbo negative conditioning stays zeroed. |
| nag_negativeopt | CONDITIONING | Unzeroed negative prompt for NAG. Fusion Rebalance/taps are applied to match the positive stream. Defaults to edit_negative_prompt in edit mode, otherwise negative. | |
| nag_phiopt | FLOAT | 4.00–20 | — |
| nag_tauopt | FLOAT | 2.500.01–20 | — |
| nag_alphaopt | FLOAT | 0.250–1 | — |
| nag_sigma_startopt | FLOAT | 1000.00–1000 | — |
| nag_sigma_endopt | FLOAT | 0.00–1000 | — |
| nag_ref_boostopt | FLOAT | 1.000–1000 | — |
| nag_ref_boost_aopt | FLOAT | 1.000–1000 | — |
| nag_fit_modeopt | COMBO | fit | 2 options: fit, crop (legacy) |
| nag_ref_boost_maskopt | MASK | — | |
| nag_auto_phiopt | BOOLEAN | false | Derive phi from alpha so alpha*phi keeps the upstream default linear guidance strength (0.25*4 = 1.0). |
| nag_phi_scaleopt | FLOAT | 1.000–4 | Multiplier for auto phi. 1.0 keeps upstream-default linear guidance strength; higher/lower scales it. |
| vae_damage_correctionopt | BOOLEAN | false | Subtract estimated VAE damage after decoding, using the selected VAE's encoder and decoder for one extra round trip per image or face crop. The 2x VAE filters back to the current image size before subtraction. No original reference is needed. |
| vae_damage_strengthopt | FLOAT | 1.000–4 | 0 skips correction; 1 is one-pass VAE damage subtraction. Values above 1 strengthen the same correction and can amplify artifacts. Does not add more iterations. |
| nag_alpha_scheduleopt | COMBO | constant | NAG alpha over this stage's executed denoising steps. The global Settings / Configuration panel mirrors the same curve to every NAG stage; each stage spans its own effective step range. |
| nag_alpha_startopt | FLOAT | 0.250–1 | First executed step's NAG alpha for the shared dynamic schedule. |
| nag_alpha_endopt | FLOAT | 0.250–1 | Last executed step's NAG alpha for the shared dynamic schedule. |
| face_prompts_jsonopt | STRING | [] | — |
| require_single_faceopt | BOOLEAN | false | — |
Outputs (6)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| cropped_refined | IMAGE | — |
| cropped_enhanced_alpha | IMAGE | — |
| mask | MASK | — |
| detailer_pipe | DETAILER_PIPE | — |
| cnet_images | IMAGE | — |