Nodes/WAS Node Suite v3/Vivid Sharpen (V2)
ComfyUI Node Runs on cloud

Vivid Sharpen (V2)

The fine-tunable version of the high-pass sharpen trick

By WASasquatch·Created 3 years ago·Updated 3 days ago· 1,832
Vivid Sharpen (V2)
  • images
  • images
radius_highpass5.00
radius_blur2.50
blur_modegaussian
hp_brightness1.00
hp_contrast1.00
vivid_opacity1.00
overlay_opacity1.00
strength1.00

This is a from-scratch, torch-only rebuild of this pack's plain VividSharpen node - same Photoshop-style high-pass-and-blend sharpening technique underneath, but with every stage of it exposed as its own control instead of one strength slider. Reach for this one when the plain version either isn't strong enough or is haloing and you want to fix the sharpening mask itself rather than just dialing strength up or down.

How it works

The core idea is unchanged from V1: build a high-pass layer and blend it back to boost perceived detail. V2 splits that into two separate blur stages instead of one - radius_highpass builds the high-pass layer itself, and radius_blur then further blurs that inverted layer before it gets blended in. Softening the sharpening mask this way is what tames the harsh, aliased edges that plain high-pass sharpening is prone to at high radii - it's the fix for exactly the haloing problem V1's article flags as a failure mode.

On top of that, hp_brightness and hp_contrast let you push more or less midtone punch into the high-pass layer before it's blended, and vivid_opacity/overlay_opacity separately weight two classic Photoshop blend modes - Vivid Light and Overlay - that stack together rather than picking one. blur_mode (gaussian or box) picks the kernel shape for both blur stages. And strength here goes up to 3.0, not V1's capped 1.0, so you can genuinely overshoot the algorithm's natural range for a more aggressive result than V1 can produce at all.

The inputs and outputs that matter

  • images (IMAGE) - what you're sharpening.
  • radius_highpass (default 5) - builds the high-pass layer; the main "how fine or broad is the detail I'm targeting" control, same role as V1's radius.
  • radius_blur (default 2.5) - softens that high-pass layer before blending; raise this if you're seeing hard, aliased sharpening edges.
  • blur_mode (gaussian / box) - kernel type for both blurs.
  • strength (default 1, range 0–3) - can exceed 1.0 for a more aggressive effect than the base algorithm's natural range.
  • hp_brightness / hp_contrast (default 1 each, 0.5–2) - tune the high-pass layer's own contrast before it's blended in. Leave these alone until the basics are dialed in.
  • vivid_opacity / overlay_opacity (default 1 each, 0–1) - independently weight the Vivid Light and Overlay blend contributions; lower one to bias the look toward the other's character.

Single output: images.

How to install it

Via ComfyUI Manager: search "WAS_Extras", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/WAS_Extras

Restart ComfyUI. No extra dependencies - it's a torch-only reimplementation, so nothing new to download beyond what ComfyUI already needs.

Common issues & troubleshooting

Not sure where to start with eight parameters. Leave hp_brightness, hp_contrast, and blur_mode at default and adjust only radius_highpass and strength first - that mirrors V1's two-knob experience while leaving the rest of V2's controls available once you need them.

V1 was haloing and you switched here expecting it to just be fixed. It isn't automatic - raise radius_blur specifically, since that's the stage that softens the sharpening mask. Simply matching V1's radius into radius_highpass without touching radius_blur will often reproduce the same haloing.

Effect looks flat despite a high strength. Check vivid_opacity and overlay_opacity haven't been turned down - both blend contributions can be independently muted, and a high strength on top of low opacities will still look weak.

Pushing strength above 1.0 makes things look broken rather than "more sharpened." That range exists for deliberate overshoot, but most images will fall apart well before 3.0 - treat anything past roughly 1.5–2.0 as an experimental extreme, not a normal working range.

CategoryWAS Suite/Image/Filter

Inputs (9)

NameTypeDefaultDescription
imagesIMAGEThe pictures to sharpen. The whole batch is processed in one pass, so a long sequence costs little more than a single frame.
radius_highpassFLOAT5.000.01–64Size in pixels of the blur that builds the high-pass layer. This sets which detail is accented: 1.0 to 2.0 for fine texture, 5.0 and above for broad shapes and general punch.
radius_blurFLOAT2.500.01–64Size in pixels of a second blur applied to the high-pass layer. Raising it softens the accent and suppresses the halo that appears along hard edges; lowering it keeps the result crisp.
blur_modeCOMBOgaussianShape of both blurs. `gaussian` falls off smoothly and is the natural-looking choice; `box` weights every pixel in the window equally, which is faster and gives a harder, more graphic accent.
hp_brightnessFLOAT1.000.5–2Brightness of the high-pass layer before it is blended. Above 1.0 pushes the result lighter overall, below 1.0 darker. Use it to correct the slight lift or drop sharpening leaves behind; 1.0 changes nothing.
hp_contrastFLOAT1.000.5–2Contrast of the high-pass layer before it is blended. Above 1.0 makes the accent bite harder, below 1.0 softens it. This is the control to reach for when the sharpening is right but too strong.
vivid_opacityFLOAT1.000–1How much of the vivid-light blend is kept. This is the stage that creates the edge accent, so 0.0 disables the sharpening and leaves only the overlay pass.
overlay_opacityFLOAT1.000–1How much of the overlay pass is kept. Overlay restores the contrast the vivid-light stage flattens, so lowering it gives a flatter, more filmic result.
strengthFLOAT1.000–3How much of the finished result is mixed back over the original. 0.0 returns the picture untouched, 1.0 uses the result outright, and values above 1.0 push past it for an exaggerated accent.

Outputs (1)

NameTypeDescription
imagesIMAGEThe sharpened pictures, same size and batch order as the input.