Resize an image

Why an online resizer usually looks worse than it should

Resizing is the most common thing anybody does to a picture and the thing most often implemented in one line. That line is almost always a call to the browser's own drawImage, and it gets two things wrong that you can see.

It mixes stored numbers rather than light. sRGB is a curve, so averaging 0 and 255 in a file gives 128, which is a good deal darker than the half-way light your eye expects. Every edge in the picture darkens slightly, and on fine detail like text or hair the whole thing goes muddy. Here the values are taken out of the curve before they are mixed and put back afterwards.

It ignores transparency. Resampling straight RGBA drags the colour of fully transparent pixels into the visible edge, which is where the grey halo around a resized cutout comes from. Multiplying the colour by the alpha first removes it completely.

On top of that, the kernel here is Lanczos-3, which is what image editors use, rather than the two-pixel average a browser reaches for. The same code resamples for the upscaler, where the difference is easiest to see at 100 percent.

Fill, fit and stretch

"Make it 1080 by 1080" has three reasonable answers, and the useful ones are not the same shape, so the page asks rather than deciding for you.

• Fill and crop gives exactly those dimensions and cuts off whatever hangs over the edges, centred. This is what a profile picture, an Instagram post or a product square wants.
• Fit inside puts the whole picture in the box, so one side comes out smaller than you asked. Nothing is cropped and nothing is padded.
• Stretch forces both numbers and distorts the picture. It is here because some upload forms demand exact dimensions, and hiding it would only send you to a worse tool.

Batch, which everyone else charges for

Drop a folder. Each file is resized in turn, shows its new dimensions as it finishes, and the lot comes back as a ZIP. There is no limit on how many and no account.

That is worth saying because it is the usual paid feature in this market: Photoroom keeps batch mode on its Pro plan, iLoveIMG caps the free tier at 30 files a task, and Pixelcut counts batch exports against a monthly allowance. None of that is necessary when the work happens on your own device, because there is no server cost to recover.

One setting matters for a batch and is off by default: enlarging. Run a folder through "1920 pixels wide" and the pictures that were already smaller should come back untouched rather than interpolated up to a size they never had. Tick the box if you do want them enlarged.

Smaller is a different job from bigger

Making a picture smaller throws pixels away, which is exactly what you want and what this page is for. Making one larger cannot add detail that was never recorded: interpolation spreads what is there over more pixels. This page does that honestly and does not pretend otherwise.

When the answer genuinely has to be bigger, the upscaler runs a model trained on photographs that draws detail it has learned, and what AI upscaling can and cannot do is honest about where that stops being a record of your picture.

Fewer pixels is usually the real fix

Most "my picture is too big" problems are about file size rather than dimensions, and resizing is the first and largest lever: halving the width removes about three quarters of the pixels. After that, the compressor takes the file down further without touching the dimensions, and reducing a PNG's file size covers the order to do them in.