| cfg | FLOAT | 1.01–30 | Classifier-Free Guidance scale.
Controls how strongly the sampler follows the prompt vs. generating freely.
Higher values = more contrast and prompt adherence, but can oversaturate.
Recommended values:
1.0 — no guidance (default, best for distilled/turbo models)
1.5-3.0 — light guidance, reduces paleness
4.0-7.0 — standard guidance for non-distilled models
8.0+ — strong guidance, risk of artifacts |
| negative_prompt | STRING | | Additional negative prompt text for inpainting.
This is encoded and combined with the negative conditioning
input on the Person Detailer node. If no negative conditioning
is connected, this text alone is used as negative.
Useful for suppressing common inpaint artifacts:
'blurry, washed out, pale, low contrast'
'extra fingers, deformed hands, bad anatomy'
Leave empty to use only the connected negative conditioning. |
| denoise_gradient | FLOAT | 0.00–1 | Center-to-edge denoise falloff during sampling.
Creates a gradient denoise mask: full denoise at the center of the
masked region, fading toward the edges.
0.0 = uniform denoise everywhere (default)
0.3 = gentle gradient (edges stay closer to original)
0.5 = moderate gradient (recommended starting point)
1.0 = full gradient (edges almost untouched by sampling)
Uses ComfyUI's native denoise_mask — the sampler blends with the
original at every step, proportional to the gradient value. |
| denoise_gradient_mode | COMBO | linear | Gradient distribution shape for denoise_gradient.
linear — USD Upscale style linear ramp from mask edge inward.
Smoothest, best for most cases. Reaches full denoise quickly.
radial — circular falloff from mask centroid outward.
Good for roughly circular masks like faces.
smooth — follows the exact mask shape using distance transform.
Best for irregular masks (body parts, accessories). |
| mask_fill_holes | BOOLEAN | true | Fill holes in masks before inpainting (closes gaps in segmentation) |
| context_expand_factor | FLOAT | 1.201–3 | Expand crop region beyond mask bbox by this factor (1.0 = tight crop) |
| output_padding | INT | 320–128 | Additional padding around crop region in pixels |
| denoise_progression | STRING | | Global per-round denoise override for ALL slots (applied when rounds > 1).
Pipe-separated values, one per round.
Examples:
0.5|0.3 — round 1: denoise 0.5, round 2: denoise 0.3
0.6|0.4|0.2 — 3 rounds with decreasing denoise
0.5|0.2 — strong first pass, gentle refinement
If empty: PersonDetailer's global denoise is used for all rounds.
If fewer values than rounds: last value repeats.
If more values than rounds: extra values are ignored. |
| steps_progression | STRING | | Global per-round steps override for ALL slots (applied when rounds > 1).
Pipe-separated values, one per round.
Examples:
6|4 — round 1: 6 steps, round 2: 4 steps
8|6|4 — 3 rounds with decreasing steps
10|4 — thorough first pass, quick refinement
If empty: PersonDetailer's global steps are used for all rounds.
If fewer values than rounds: last value repeats.
If more values than rounds: extra values are ignored. |
| reference_1_ref_index | INT | 11–10 | Which reference (1-10) this row configures.
Default: row 1 → reference 1.
Person Selector supports up to 10 references. |
| reference_1_mask_type | COMBO | head | Mask type for this row.
face=skin only, head=face+hair, body=full body,
aux=body-part SEGS from detector |
| reference_1_rounds | INT | 11–10 | Number of inpaint passes for this row.
Each round re-encodes the previous result (latent cycling).
Use denoise_progression/steps_progression to control per-round values.
1 = single pass (default), 2-3 = progressive refinement |
| reference_1_detail_daemon | BOOLEAN | true | Enable Detail Daemon sigma manipulation for this row.
Enhances detail preservation during inpainting. |
| reference_2_ref_index | INT | 21–10 | Which reference (1-10) this row configures.
Default: row 2 → reference 2.
Person Selector supports up to 10 references. |
| reference_2_mask_type | COMBO | head | Mask type for this row.
face=skin only, head=face+hair, body=full body,
aux=body-part SEGS from detector |
| reference_2_rounds | INT | 11–10 | Number of inpaint passes for this row.
Each round re-encodes the previous result (latent cycling).
Use denoise_progression/steps_progression to control per-round values.
1 = single pass (default), 2-3 = progressive refinement |
| reference_2_detail_daemon | BOOLEAN | true | Enable Detail Daemon sigma manipulation for this row.
Enhances detail preservation during inpainting. |
| reference_3_ref_index | INT | 31–10 | Which reference (1-10) this row configures.
Default: row 3 → reference 3.
Person Selector supports up to 10 references. |
| reference_3_mask_type | COMBO | head | Mask type for this row.
face=skin only, head=face+hair, body=full body,
aux=body-part SEGS from detector |
| reference_3_rounds | INT | 11–10 | Number of inpaint passes for this row.
Each round re-encodes the previous result (latent cycling).
Use denoise_progression/steps_progression to control per-round values.
1 = single pass (default), 2-3 = progressive refinement |
| reference_3_detail_daemon | BOOLEAN | true | Enable Detail Daemon sigma manipulation for this row.
Enhances detail preservation during inpainting. |
| reference_4_ref_index | INT | 41–10 | Which reference (1-10) this row configures.
Default: row 4 → reference 4.
Person Selector supports up to 10 references. |
| reference_4_mask_type | COMBO | head | Mask type for this row.
face=skin only, head=face+hair, body=full body,
aux=body-part SEGS from detector |
| reference_4_rounds | INT | 11–10 | Number of inpaint passes for this row.
Each round re-encodes the previous result (latent cycling).
Use denoise_progression/steps_progression to control per-round values.
1 = single pass (default), 2-3 = progressive refinement |
| reference_4_detail_daemon | BOOLEAN | true | Enable Detail Daemon sigma manipulation for this row.
Enhances detail preservation during inpainting. |
| reference_5_ref_index | INT | 51–10 | Which reference (1-10) this row configures.
Default: row 5 → reference 5.
Person Selector supports up to 10 references. |
| reference_5_mask_type | COMBO | head | Mask type for this row.
face=skin only, head=face+hair, body=full body,
aux=body-part SEGS from detector |
| reference_5_rounds | INT | 11–10 | Number of inpaint passes for this row.
Each round re-encodes the previous result (latent cycling).
Use denoise_progression/steps_progression to control per-round values.
1 = single pass (default), 2-3 = progressive refinement |
| reference_5_detail_daemon | BOOLEAN | true | Enable Detail Daemon sigma manipulation for this row.
Enhances detail preservation during inpainting. |
| generic_mask_type | COMBO | head | Mask type for generic (unmatched) slot.
face=skin only, head=face+hair, body=full body,
aux=body-part SEGS from detector |
| generic_rounds | INT | 11–10 | Number of inpaint passes for generic slot.
Each round re-encodes the previous result (latent cycling).
Use denoise_progression/steps_progression to control per-round values.
1 = single pass (default), 2-3 = progressive refinement |
| generic_detail_daemon | BOOLEAN | true | Enable Detail Daemon sigma manipulation for generic slot.
Enhances detail preservation during inpainting. |