SharpnessPro · Netteté optimisée
The sharpening node you should reach for first
- image
- IMAGE
You know the look. The image comes out of the sampler, you upscale it, and everything is just… soft. Not blurry-broken, just lacking that final crispness. Most "sharpen" nodes in ComfyUI are crude: they boost the difference against a plain gaussian blur and call it a day, which is how you get halos and crunchy noise. SmartSharpen is the pack's flagship for a reason - it's the one the author tells you to use by default, and unlike the usual sharpen-at-any-cost crowd, it actually tries not to wreck your image while doing it.
How it works
The trick is that the blur it compares against isn't a plain gaussian. SmartSharpen uses an edge-aware smooth (a lightweight guided filter) on the luminance channel, so the "detail" it extracts skips over real edges instead of amplifying them into white lines. From there it does three protective things in one pass:
- reduce_noise compresses tiny high-frequency differences so grain and JPEG-ish noise don't get boosted into sand.
- fade_shadows and fade_highlights mask the effect out of dark and bright zones, which is where sharpening normally clips to black/white and grows halos.
- Everything is recomputed on luminance and recombined preserving chroma, so you don't get color fringes on edges.
That's the whole reason to pick it over the other nodes here: it sharpens the midtone detail and leaves the extremes alone.
The inputs that matter
Two sliders do the heavy lifting, and the rest are damage control:
- radius (default 1.8) - how big the details you're boosting are. Low radius targets fine edges; crank it up and you're sharpening broader structure, which gets mushy fast.
- amount (default 1.2) - how hard the detail is applied. 1.0 is a light hand; past ~2.5 you're in halo territory.
- reduce_noise (default 0.15) - raise this if the source is noisy; it's the reason SmartSharpen survives low-light renders that would wreck a plain USM.
- fade_shadows / fade_highlights (default 0.2 each) - protection for dark and bright areas. If you see white halos on highlights or grit in shadows, bump these before you touch amount.
It takes an IMAGE and outputs an IMAGE, so it drops anywhere in a graph: wire it after upscaling, right before Save Image. All the sharpen nodes in this pack are pure PyTorch (no models, no downloads), so it's fast even on CPU.
Install
Grab it with ComfyUI Manager - search SharpnessPro under Install Custom Nodes - or clone it manually:
cd ComfyUI/custom_nodes/
git clone https://github.com/orion4d/ComfyUI_SharpnessPro.git
Restart ComfyUI. That's it: no requirements.txt, no model files, nothing heavy.
Common issues
Over-sharpening is the one failure mode that actually bites people. The defaults are sane, but the temptation to push amount to 3 is real. Zoom to 100% and check the edges; if you see white halos, back off amount and raise fade_highlights rather than fighting it. The pack's own advice applies here more than anywhere: start gentle, because you can always sharpen more, but you can't un-sharpen a crunchy render.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| radius | FLOAT | 1.80–50 | — |
| amount | FLOAT | 1.200–5 | — |
| reduce_noise | FLOAT | 0.150–1 | — |
| fade_shadows | FLOAT | 0.200–1 | — |
| fade_highlights | FLOAT | 0.200–1 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |