Category: Image tools
Sharpen Image Online
A fixed 3x3 unsharp-mask convolution, one strength control, PNG out
One 0-100 strength slider drives a fixed 3x3 unsharp-mask convolution in your browser, then downloads a PNG. Edge contrast only, not AI unblur. Nothing is uploaded.
Sharpening emphasizes edges using an unsharp mask; high strength can amplify noise, so start low.
Everything on this page is processed in your browser. Nothing is uploaded.
Everything on this page is processed in your browser. Nothing is uploaded.
What this tool does
This tool has exactly one control: a strength slider from 0 to 100. Behind it sits a fixed 3x3 unsharp-mask convolution, run in JavaScript over the canvas pixel data — the centre pixel is weighted by one plus four times the amount, and the four orthogonal neighbours each by the negative amount. The slider maps onto a convolution amount from 0 to 1.6, so the top of the range is a genuinely strong sharpen rather than a token one, and at strength 0 the convolution is skipped entirely and your pixels are untouched. At the frame edges the sampling coordinates are clamped to the nearest in-bounds pixel, so the border picks up no artefact from reading outside the image. Only the three colour channels are sharpened; the alpha channel is copied through unchanged, so transparency survives exactly. Results are held in a clamped byte array, which means an over-strong setting flattens highlights instead of wrapping colours around. Slider movement is debounced by 120 milliseconds so dragging does not queue a full-image convolution per step. The working canvas is scaled so the longest side is at most 1200 pixels — a smaller image is never enlarged — and the output is PNG, which avoids stacking fresh compression artefacts on top of freshly sharpened edges. Be clear about what this is not. It is not AI: no model is downloaded or run, there is no neural unblur, enhancement or upscaling. Sharpening raises edge contrast and apparent sharpness; it cannot recover detail that is not in the file, so motion blur, a missed focus and heavy compression damage are not undone. It will also amplify whatever grain is already present, and there is no noise-aware behaviour and no denoising stage here. There is no full unsharp mask with amount, radius and threshold — radius and threshold are fixed; the kernel is not configurable and never samples the diagonals; there is no edge-aware, frequency-separated or luminance-only mode and no deconvolution; there is no masking or selective application, no before-and-after or reset, no format or quality choice, and no batch. If your browser cannot decode the file, nothing is reported and nothing appears. Your photo is never uploaded to LocalTools.
How to use it
- Choose one image; it is decoded once and kept in memory, and slider moves are debounced by 120 ms.
- Set the strength between 0 and 100 — 0 leaves the pixels untouched, the top of the range is a strong 1.6 convolution amount.
- Check the edges for halos, then download the PNG; the working canvas caps the longest side at 1200 pixels.
Privacy
This tool runs entirely in your browser. Your input is never uploaded, stored or shared — closing the tab removes it.
Frequently asked questions
- How do I sharpen an image here?
- Choose a photo and drag the single strength slider. Around a quarter to a half of the range suits a slightly soft shot; the preview and the downloadable PNG update together as you move it.
- Can sharpening fix a blurry photo?
- Only mild softness. This is a 3x3 convolution that raises edge contrast, so it cannot recover severe out-of-focus or motion blur — that detail is simply not in the file, and there is no AI model here to invent it.
- Can sharpening introduce artifacts?
- Yes. Strong settings create bright halos along high-contrast edges and amplify any grain already in the photo. There is no noise-aware behaviour and no denoising stage, so back the strength off when halos appear.
- Is this a real unsharp mask with amount, radius and threshold?
- It is an unsharp-mask style 3x3 convolution, but only the amount is adjustable: radius and threshold are fixed, the kernel samples the four orthogonal neighbours and not the diagonals, and it cannot be configured.
- Does the size or quality change after sharpening?
- The longest side is capped at 1200 pixels, so a larger photo comes back smaller; a smaller one is never enlarged. Output is always PNG, with no format or quality choice, and there is no batch mode.