ComfyUI Node
π§Ή Advanced Image Denoiser
Edge-preserving denoising with automatic noise estimation and edge-aware detail recovery β removes noise without blur.
π§Ή Advanced Image Denoiser
- image
- image
- noise_report
βmethodsmart_autoβΊ
βstrength0.15βΊ
βdetail_recovery0.35βΊ
βluminance_strength0.10βΊ
βchroma_strength0.30βΊ
βpatch_size7βΊ
βsearch_window21βΊ
βwavelet_level3βΊ
βblend_original0.00βΊ
βsharpen_modeoffβΊ
βsharpen_amount0.20βΊ
βsharpen_radius0.30βΊ
Categoryimage/denoising
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | β | |
| method | COMBO | smart_auto | smart_auto measures the noise level and applies just enough denoising. bm3d is the highest quality (needs `pip install bm3d`). |
| strength | FLOAT | 0.150β1 | Keep this LOW (0.05β0.25). In smart_auto, 0.5 applies exactly the measured noise level; higher over-smooths. |
| detail_recovery | FLOAT | 0.350β1 | Edge-aware: restores fine texture from the original along edges only, so flat areas stay clean. Safe to raise. |
| luminance_strengthopt | FLOAT | 0.100β1 | Manual methods only. Brightness-channel denoising β keep low, this is what causes blur if overdone. |
| chroma_strengthopt | FLOAT | 0.300β1 | Color-noise removal. Eyes are less sensitive to chroma blur, so this can be 2β3Γ higher than luminance. |
| patch_sizeopt | INT | 73β15 | NLM comparison patch (odd). 7 is standard. |
| search_windowopt | INT | 217β35 | NLM search area (odd). Bigger = better but slower. |
| wavelet_levelopt | INT | 31β6 | Wavelet decomposition depth. |
| blend_originalopt | FLOAT | 0.000β1 | Mix the untouched original back in for a natural look (0.1β0.2 is plenty). |
| sharpen_modeopt | COMBO | off | Optional post-sharpen. luminance_only avoids color fringing. |
| sharpen_amountopt | FLOAT | 0.200β1 | Sharpening intensity. |
| sharpen_radiusopt | FLOAT | 0.300.05β1 | Detail scale: low = fine micro-detail. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | β |
| noise_report | STRING | β |