ComfyUI Node
FeiHou Easy H3 Face Refine (Experimental)
Experimental H3 face crop/refine/stitch. Adds sampling; does not guarantee lip sync. Original audio passes through.
FeiHou Easy H3 Face Refine (Experimental)
- images
- model
- h3_context
- reference_face
- audio
- images
- audio
- fps
- report
◄enabledtrue►
◄detector▾►
◄target▾►
◄denoise0.25►
◄steps8►
◄seed0►
◄advancedfalse►
◄canvas_size768►
◄confidence0.35►
◄crop_factor2.5►
◄chunk_frames不分段►
◄sampler_nameeuler►
◄schedulersimple►
◄promptRefine the face in this video crop. Preserve identity, expression, gaze, mouth movement, lighting and motion. Natural facial detail; no new motion or camera changes.►
◄force_offloadtrue►
◄audio_lockfalse►
◄clean_second_modeltrue►
CategoryFeiHou Easy H3
Inputs (22)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | — | |
| model | MODEL | — | |
| h3_context | MINIMAX_H3_CONTEXT | — | |
| enabled | BOOLEAN | true | — |
| detector | COMBO | 1 options: face_yolov8m.pt | |
| target | COMBO | 2 options: largest_face, reference_identity | |
| denoise | FLOAT | 0.250–1 | — |
| steps | INT | 81–100 | — |
| seed | INT | 00–18446744073709550000 | — |
| advanced | BOOLEAN | false | — |
| canvas_size | COMBO | 768 | 2 options: 512, 768 |
| confidence | FLOAT | 0.350.05–0.95 | — |
| crop_factor | FLOAT | 2.51.2–5 | — |
| chunk_frames | COMBO | 不分段 | 5 options: 不分段, 240, 192, 120, 72 |
| sampler_name | COMBO | euler | 44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38 |
| scheduler | COMBO | simple | 9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3 |
| prompt | STRING | Refine the face in this video crop. Preserve identity, expression, gaze, mouth movement, lighting and motion. Natural facial detail; no new motion or camera changes. | — |
| force_offload | BOOLEAN | true | — |
| audio_lock | BOOLEAN | false | Experimental: condition face sampling on the connected, time-aligned original audio. Not a guaranteed lip-sync correction. |
| clean_second_model | BOOLEAN | true | Use the recorded pre-LoRA second-pass base. Requires an Easy H3 second-pass model; removes external patches as well. |
| reference_faceopt | IMAGE | — | |
| audioopt | AUDIO | — |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| audio | AUDIO | — |
| fps | FLOAT | — |
| report | STRING | — |