Resize an Image Without Losing Quality: Down vs Up
You can resize an image without losing quality in one direction only: down. Making a picture smaller throws away pixels you no longer need, and done once, with a sensible save setting, nobody can tell. Making it bigger is different. The detail isn't in the file, so the software invents it, and invented detail looks soft.
That's the whole story in two sentences. The rest of this page is the numbers behind it, plus the three habits that cause most of the damage people blame on resizing.
Short answer: what to do in each case
Going smaller: keep the aspect ratio locked, resize once from the original, save as JPEG at quality 80 to 85 (or WebP), and stop. Going bigger: don't, if there's any way to get a larger original. If there isn't, stay under about 1.5 times the original width and expect a slightly soft result.
Simple enough. Here's why.
Scaling down: what actually gets lost
A 4000 px wide phone photo shown in a 1200 px wide blog column is carrying eleven times more pixels than the screen can use. Resizing to 1200 px merges neighbouring pixels into averages. Fine detail smaller than the new pixel size disappears, but it was never going to be visible at that display size anyway.
I ran a quick test on four detailed images, each 1920 px wide (two interiors, two exteriors with foliage and brick). All sizes below are JPEG at quality 80.
| Width | Total size, 4 images | Change |
|---|---|---|
| 1920 px (original) | 1,144 KB | baseline |
| 1600 px | 866 KB | 24% smaller |
| 800 px | 254 KB | 78% smaller |
Halving the width from 1600 to 800 px leaves a quarter of the pixels and cut the files by 71%. That's why resizing beats every compression trick for getting file size down: you remove data instead of squeezing it. If a file is still too heavy after that, the Image Compressor handles the last stretch.
The method used to shrink matters a little. Good resizers average pixels (names you may see: bicubic, Lanczos, "high quality"). The crude method, nearest neighbour, just drops rows and columns, which leaves jagged edges on diagonal lines and a shimmer on fine patterns like brick or fabric. In my test the crude method scored 35.7 dB against a proper downscale, a visible difference on roof lines and window frames. Browser-based tools that draw to a canvas with smoothing on, and every mainstream photo editor, use an averaging method. You'd have to go out of your way to get the bad one.
Scaling up: you can't add pixels that were never there
Now the other direction. I took the 800 px versions, enlarged them back to 1600 px, and compared the result with the real 1600 px images using PSNR, a similarity score where higher means closer to the reference. Above 40 dB or so, differences are hard to find. In the low 30s, softness is plain to see.
| Enlarging method (800 to 1600 px) | Similarity to the real 1600 px image |
|---|---|
| Nearest neighbour | 31.2 dB |
| Bilinear | 32.2 dB |
| Bicubic | 33.3 dB |
| Lanczos | 33.8 dB |
| For comparison: real 1600 px image saved at JPEG quality 80 | 38.5 dB |
Two things stand out.
The best and worst enlarging methods are only 2.6 dB apart, so no setting rescues an upscale. And every one of them is far below an ordinary quality 80 save. Doubling an image does more damage than the JPEG compression people worry about.
It gets worse on file size: the four enlarged images totalled 700 KB, against 254 KB for the 800 px files they were made from, which is 2.75 times the bytes for zero extra detail.
What does a 2x upscale look like in practice? Hair and grass turn to a smooth paste. Text grows a faint halo. Edges that were one pixel wide become a three-pixel blur. At 1.25x or 1.5x the same thing happens, but mildly enough that most photos get away with it, especially on a phone screen.
Some newer upscalers use trained models to guess plausible detail, and they can look sharper than the methods above. They're still guessing. For a face, a product label or anything with small text, a guess can be wrong in ways that matter, so I'd treat them as a last resort for decorative images and never for documents or ID photos.
Three mistakes that get blamed on resizing
Stretching instead of cropping. If you type 1080 x 1080 into a resizer for a 1920 x 1080 photo with the aspect ratio lock switched off, everything gets squashed to 56% of its width. People look thin, circles turn into ovals. That isn't quality loss, it's distortion, and it's the most common resizing trap. The fix is order: cut the picture to the target shape first with the Image Cropper, then resize the crop.
Resizing a copy of a copy. Each JPEG save loses a little, and resizing an already-resized file stacks the losses. I tried it: shrink to 1500 px, save, enlarge back to 1600 px, repeated ten times. The result scored 36.9 dB against the starting image, noticeably softer in fine textures. One careless round doesn't ruin a photo. Ten do.
Always go back to the original file.
Saving at quality 100 "to be safe". Quality 100 doesn't protect a resized image, it just makes the file two to three times larger than quality 90 with nothing to show for it. The 80 vs 90 vs 100 comparison has the file sizes.
Screenshots, logos and text are a different case
Everything above is about photos. Screenshots, logos, charts and scanned text have hard edges and flat colours, and they behave differently in two ways.
First, they shrink badly at odd ratios. A screenshot reduced to 73% turns one-pixel lines into grey smears, because a line can't be 0.73 pixels wide. Clean ratios (exactly 50%, or 25%) hold up far better, so if you control the size, halve it.
Second, save them as PNG, not JPEG. JPEG puts a fuzzy ring around sharp text, and that ring is usually what people mean when they say the resize "lost quality". With a logo, the real answer is to ask for the vector file (SVG, or a PDF from the designer), which scales to any size with no loss at all.
How big does the image need to be?
Resize to what the destination shows, with a margin for sharp screens. A blog column 800 px wide looks crisp with a 1600 px image on a high-density display and perfectly fine with 1200 px. A full-width banner rarely needs more than 1920 to 2400 px. For email, 1200 to 1600 px on the long side is plenty. Print is the one place that needs the big file: at 300 dots per inch, a 10 x 15 cm print wants about 1200 x 1800 px, and an A4 page about 2480 x 3508 px.
If you don't know the target, the image size cheat sheet lists common web, social and email dimensions. And if the destination sets a limit in KB instead of pixels, resize first and compress second. Trying to hit 200 KB on a 4000 px photo by quality alone means dropping to settings where blocks appear, while the same photo at 1600 px gets there at quality 80 without any visible harm.
What I'd do, step by step
- Start from the original file, never from a version that was sent through a chat app or downloaded from a social site (those have already been shrunk and recompressed).
- Crop to the shape you need.
- Resize once, aspect ratio locked, to the largest size the destination will display.
- Save photos as JPEG at 80 to 85 or as WebP; save screenshots and graphics as PNG.
- Keep the original. Next time you need a different size, repeat from step one.
The Image Resizer does steps three and four in the browser: type a width or a height in pixels, or pick a percentage, and leave Maintain Aspect Ratio ticked so nothing stretches. Files are processed on your device and aren't uploaded, and there's no account to create. One photo takes about as long as reading this sentence.
Frequently asked questions
Does resizing an image reduce its quality?
Making it smaller removes detail that the smaller size couldn't show anyway, so the result looks clean. Making it larger always softens it, because the extra pixels are estimated from their neighbours. Quality also drops if the proportions are changed or the file is re-saved many times.
How do I make a small image bigger without it going blurry?
Find a larger original if one exists: the camera file, the sender's uncompressed version, or the vector file for a logo. If none exists, enlarge by no more than about 1.5 times and accept slight softness. A 2x enlargement is visibly blurred with any standard method.
Should I resize or compress first?
Resize first. Reducing pixel dimensions removes the most data for the least visible change. Compress afterwards, once, at a moderate quality setting. Doing it the other way round means compressing pixels you are about to throw away.
Why does my resized picture look stretched?
The width and height were set independently, so the shape changed. Lock the aspect ratio and enter only one dimension, or crop the image to the target shape before resizing it.