01 / Drop
Choose a picture
A blurry photo, a soft screenshot or a picture that was saved too small.
Free tool
Give a soft photo, a screenshot or a scan crisp edges. A plain sharpen filter only adds contrast to edges and makes noise worse; this one uses an open AI model, Real-ESRGAN, that rebuilds the edges and cleans compression marks. It runs in this tab, so nothing is uploaded. No sign-up, no watermark.
Pick a stronger setting if the result is still soft, a weaker one if it looks painted. Lower the AI strength if edges look too hard.
This runs in your browser. Your pictures are never uploaded.
How it works
01 / Drop
A blurry photo, a soft screenshot or a picture that was saved too small.
02 / Sharpen
The first time, the page loads the AI model, under 3 MB. Then your graphics card does the work.
03 / Compare
Drag the line to compare with the original, change the strength if you like, then save it at its own size.
Questions
No. The AI model is downloaded to your browser and runs there, so the picture never leaves your computer.
A sharpen filter, like the one in a photo app, raises the contrast on both sides of every edge. It makes a soft picture look crisper, but it also sharpens noise and JPG blocks. The AI rebuilds the picture instead: edges get sharper and the noise and blocks get cleaner.
Real-ESRGAN General x4v3, an open model by Xintao Wang and others under the BSD licence, trained to repair soft, small and compressed pictures. The page first makes the picture smaller, which shrinks the blur, lets the model rebuild it four times larger, and saves it at its original size. The stronger the setting, the smaller it is made first.
No. It guesses detail that is not in the picture. Text that is soft but readable usually comes out crisp; text you cannot read at all can come out as made-up shapes, so never trust it for names, numbers or evidence.
Only a little. It is trained on soft and out of focus pictures; strong streaks from a moving camera stay. It also cannot bring back a face that is only a few pixels wide.
At 100% you get the AI picture alone. Lower values mix it with your original, which softens edges that look too hard or a face that looks painted.
It needs WebGL, which current Chrome, Edge, Safari and Firefox have. Without it, the page runs the model on the processor, which is much slower.