ComfyUI Node
LC Smart Detailer (pipe) 🩹
Finds what you describe (hands, face, eyes, etc.) and redraws it at full detail. Each hand or face gets its own crop, scaled up to inpaint_resolution, redrawn, then blended back in. Nothing found? The image passes through untouched and the node tells you. No negative wired: fine for Flux, Krea 2, Anima and other cfg 1 models. Wire one for SDXL.
LC Smart Detailer (pipe) 🩹
- pipe
- image
- model
- vae
- positive
- inpaint_positive
- negative
- pipe
- image
- mask
◄prompthands►
◄sam3_model▾►
◄threshold0.50►
◄grow4.0►
◄feather8.0►
◄blend0.80►
◄padding64►
◄inpaint_resolution1024►
◄denoise0.30►
CategoryLC MaskMaker/detailer
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| pipe | LC_PIPE | LC pipe: model, VAE, seed, steps, cfg, sampler and scheduler come from here. | |
| image | IMAGE | — | |
| prompt | STRING | hands | What to detail, as simple words separated by commas: hands, face. Every match is found (both hands, every face). Add :N to cap it, e.g. face:1. |
| sam3_model | COMBO | SAM 3 model that does the finding. Pick the SAM 3.1 download, or use your own sam3.pt / sam3.safetensors in models/sam3. | |
| threshold | FLOAT | 0.500.05–0.95 | How sure SAM 3 has to be. Lower finds more (and more wrong things), higher is pickier. 0.5 is a good start. Colors are only a hint: 'red dress' can still pick a green dress. Things that are not there are not found. |
| grow | FLOAT | 4.0-64–256 | Pixels to grow (positive) or shrink (negative) the mask before inpainting. Growing a little covers the edges. |
| feather | FLOAT | 8.00–128 | Softness of the mask edge in pixels, so the redrawn area blends into the rest. 0 = hard edge. |
| blend | FLOAT | 0.800–1 | How much of the redraw is pasted in. 1 = all of it, 0.5 = half redraw, half original. Lower tames a redraw that went too far. |
| padding | INT | 640–1024 | Pixels of the surrounding image included with each crop, so the model sees the context. |
| inpaint_resolution | INT | 1024256–4096 | Each crop is redrawn at about this many pixels squared (1024 = 1 megapixel), whatever its shape. Higher = more detail, slower. |
| denoise | FLOAT | 0.300–1 | How much is redrawn. Low keeps the shape and fixes details, high redraws from scratch. |
| modelopt | MODEL | Optional. Wire a model here to use it instead of the pipe's (with LoRAs on it, or a different model). | |
| vaeopt | VAE | Optional. Wire a VAE here to use it instead of the pipe's. | |
| positiveopt | CONDITIONING | The image's prompt. A prompt describing the whole scene can get drawn into every crop at denoise 0.3 and up (a whole tiny scene in the face): wire inpaint_positive to prevent that. | |
| inpaint_positiveopt | CONDITIONING | Optional. A prompt for the crops only, e.g. 'close-up of a woman's face, natural skin'. Used instead of positive for the redraw. Recommended whenever positive describes the whole scene. | |
| negativeopt | CONDITIONING | Optional. Leave empty for cfg 1 models; wire it for SDXL and other models that use a negative. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| pipe | LC_PIPE | — |
| image | IMAGE | — |
| mask | MASK | — |