How to Resize Images Without Losing Quality

"Resize this but don't lose quality" sounds like a single request, but it's really two very different problems. Making an image smaller is almost free. Making it bigger fights physics. Knowing which one you're doing is most of the battle.

What actually happens when you resize

A raster image is a fixed grid of pixels. Resizing throws that grid away and paints a new one at a different resolution: a process called resampling. The software has to decide what colour each new pixel should be based on the old ones nearby.

• Downscaling merges many source pixels into one. There's plenty of information to average, so results look clean.
• Upscaling stretches few source pixels across many. The software must invent the in-between values, which is where softness and blur creep in.

That asymmetry explains almost everything about resize quality.

Shrinking: keep it crisp

Downscaling is where you can win easily.

1. Resize once, at the end. Repeatedly nudging the size down compounds softening. Go from the original straight to the final dimensions.
2. Use a high-quality resampler. The filter matters more than people think. A crude "nearest pixel" grab produces jagged, aliased diagonals. Simple bilinear smoothing is better. A windowed filter like Lanczos (which weighs a wider neighbourhood of source pixels) keeps edges clean and sharp.
3. Re-sharpen lightly if needed. A tiny sharpen after a big downscale restores the perceived crispness that averaging softened.

For what it's worth, This is what our own image resizer does, since I get asked: it resamples with the pica library's Lanczos filter, the same algorithm serious desktop editors use, and falls back to the browser's built-in high-quality canvas smoothing only if that can't load. Either way, a straight shrink comes out clean with no extra steps, and the image never leaves your device.

Enlarging: manage expectations

No resizer can show detail the camera never captured. What you can do is enlarge gracefully:

• Keep enlargements modest. Going 10–20% bigger is often invisible. Doubling is noticeable. Quadrupling is obvious.
• Use the right tool for big jumps. Past about 2x, plain resampling visibly softens. That's the point of a dedicated AI upscaler, which predicts plausible detail instead of only stretching pixels, I've written up how that works and when it helps.
• Start from the highest-resolution original you have. Enlarging an already-small or already-compressed image stacks softness on softness.

The quiet quality killer: re-compression

Even a perfect resize can look bad if you save it wrong. Every time you re-save a JPEG, it re-compresses and loses a little more. To protect quality:

• Resize from the original file, not a copy that's already been through several saves.
• Export to PNG (lossless) when you need pixel-perfect edges, logos, screenshots, text.
• Export to JPEG or WebP for photos, and don't push the quality slider too low. (Unsure which? See lossy vs lossless formats explained.)

A reliable resize checklist

1. Open the original, highest-quality file you have.
2. Decide: are you shrinking (easy) or enlarging (be conservative)?
3. Lock the aspect ratio so you don't stretch the image.
4. Resize in one step to the exact target size.
5. Choose the right output format: lossless for graphics, lossy for photos.
6. Preview at 100% before you commit.

Do it in the browser

You don't need heavy software for any of this. Drop an image into the image resizer, keep the aspect lock on, type your target size, and download. Because it runs on your device, there's no upload, no queue, and no extra round of server-side compression quietly degrading your file.

Resize smart, and "smaller" stays sharp while "bigger" stays believable.

Sources

• Lanczos resampling (Wikipedia)
• pica (high-quality in-browser image resizing library)