gizmobench

Image Sharpen

Choose a photo, drop one onto the panel or paste a screenshot, and it is sharpened here in the browser tab by an unsharp mask with all three of its controls exposed: radius, in pixels, is how wide the blur under the mask is, amount, as a percentage, is how much of the edge detail is added back, and threshold, in levels out of 255, is the smallest difference worth touching, which is what keeps flat areas like sky and skin from getting noisier. A single slider called Sharpen hides two of those three, and that is why sharpened photos so often come back with pale rims along every hard edge. Drag the divider across the picture to compare the two versions in one frame: the panel fits the photo to the room it has, while the mask itself and the download run at the full resolution of the file, up to 30 megapixels.

no image yet

or drop one here, or paste a screenshot

Never uploadedTransparency keptNo accountUp to 30 megapixels
Radius
1.2 px
Amount
85 %
Threshold
4
Save as
  • One edge, amount 8560 against 200, radius 1, threshold 4
    becomes 24 against 236, a step of 212 instead of 140
  • The same edge, amount 300the overshoot runs off both ends
    0 against 255: a flat rim, which is what a halo is
  • A cut-out on transparencyflat colour, alpha 0 around it
    every pixel identical, no rim along the alpha edge

Those three lines are printed by running the same code your picture goes through, not typed in by hand.

Your image stays on this device. An unsharp mask raises contrast along edges that are already in the image. It cannot recover detail the camera never recorded, so a blurred or heavily compressed photo cannot be made sharp, and a high amount produces bright halos along hard edges rather than more detail. Images up to 30 megapixels are accepted. There is no upload, no account and no watermark, and no image byte is handed to analytics either: the only thing the page records is that the tool was used.

Common questions

What do radius, amount and threshold actually do?
An unsharp mask blurs a copy of your photo, subtracts that blur from the original to find the local detail, and adds a multiple of that detail back. Radius is the width of that blur, as the standard deviation of a Gaussian, from 0.1 to 10 pixels: small values sharpen fine texture, large values thicken edges instead. Amount is how much of the detail is added back, from 0 to 300 percent, where 100 adds it back exactly once and 0 leaves every pixel as it was. Threshold, from 0 to 255 on the same scale the colour values are stored on, is the smallest difference between a pixel and its blurred neighbourhood that will be sharpened at all: below it, the pixel is left alone. The page starts at radius 1.2, amount 85 and threshold 4, which is a light pass for a slightly soft photo.
How do I sharpen a photo without halos?
A halo is the mask overshooting: the light side of an edge is pushed past white and the dark side past black, so a flat rim appears instead of detail. Three things prevent it. Keep the amount moderate, somewhere between 50 and 120 percent for most photographs; the page says so on screen once you pass 150. Keep the radius small, near 1 pixel for a web-sized image, because a wide radius spreads the overshoot into a visible band. And raise the threshold to 3 or 4 so that only real edges are touched and flat areas keep the values they had. The worked examples under the tool show the same edge at amount 85 and at amount 300, and the numbers come from running the code, not from an illustration.
Can a blurry photo be fixed?
No, and it is worth being plain about it. Sharpening raises contrast along edges that are already recorded in the file; it has no way to know what was in front of the lens where the sensor recorded a smooth gradient. A photo that missed focus, that was taken while the camera moved, or that has been compressed until its detail is gone will come back with more contrast and more visible artefacts, not with the detail restored. This tool does no upscaling, no noise removal and no generated detail of any kind. If the picture is soft because it was enlarged, resizing it back down often does more than any amount of sharpening.
Is transparency preserved?
Yes, in two senses. The alpha channel is copied through untouched, so a pixel that was fully transparent is still fully transparent and a half-transparent edge keeps its exact value. Beyond that, the blur under the mask is weighted by alpha, so transparent pixels contribute nothing to their neighbours. That second part matters: a sharpener that blurs straight through a cut-out reads the black that usually sits under a transparent pixel, and hands the picture back with a bright rim around the shape. Save as PNG to keep all of it. JPEG has no transparency at all, so if you pick it the page tells you before the download that those pixels will be filled with white.
Is my photo uploaded?
No. The file is read by the page itself, decoded by your browser, drawn onto a canvas and sharpened in memory. There is no server in this and no endpoint for an image to go to, nothing about the picture is sent to analytics, and nothing about it is written into your browser's storage: the tool remembers your radius, amount and threshold between visits, and that is all it remembers.
How large an image can it sharpen, and what comes out?
Up to 30 megapixels, which covers a 6,000 by 5,000 frame. A larger file is refused in words, with the megapixel size it read and the ceiling, rather than being quietly shrunk to fit; resize it first and sharpen the smaller copy. What comes out has exactly the same pixel dimensions as what went in, because sharpening changes values and not geometry. The download is your own file name with -sharpened on the end, as a PNG, or as a JPEG written at 0.92 quality when you choose that instead.

An unsharp mask raises contrast along edges that are already in the image. It cannot recover detail the camera never recorded, so a blurred or heavily compressed photo cannot be made sharp, and a high amount produces bright halos along hard edges rather than more detail. Images up to 30 megapixels are accepted.