Sharpen Image

No. An unsharp mask increases the local contrast at edges that are already in the file. If the lens never resolved the detail, there is no edge to enhance, and pushing the amount up only produces halos around the soft shapes. Deconvolution and machine-learning deblurring attack that problem, and neither is what this does.

Protecting flat areas. A photo's noise is a small pixel-to-pixel difference, and sharpening amplifies it as happily as it amplifies an eyelash. The threshold sets a minimum difference below which a pixel is left untouched, so sky, skin, and shadow stay smooth while real edges still get sharpened. Values of 3 to 10 are typical for a noisy image.

Look at a high-contrast edge, such as a branch against a bright sky, at 100%. If you can see a bright outline hugging the light side and a dark one on the other, that is a halo, and it is the signature of too much amount. Back off until it disappears. Halos are far more visible in print and on large screens than on the preview you are judging from.

After. Downscaling averages pixels together, which softens edges and also destroys any sharpening you applied at the larger size. The professional order is edit, crop, resize to final dimensions, then sharpen for the output. That is also why this tool sharpens at the 2000px working size rather than at full resolution.

Because PNG is lossless, and JPEG compression is particularly unkind to freshly sharpened edges: the same high-frequency detail you have added is what JPEG's quantisation throws away first, and it also produces ringing around it. Sharpen, save as PNG, and convert to JPEG once at the end if you need to.

No. The convolution runs pixel by pixel on an HTML canvas inside the page. The file never leaves your device, which also means large images are limited by your machine's speed rather than by a server.