How to compress an image without losing quality, measured
You cannot make a photo much smaller without discarding data, but you can discard none you would notice. On five photos, JPEG quality 82 scored 0.970 or higher for similarity, where 1 is identical. Quality 100 made files 5 to 9 times larger. The truly lossless routes saved under 1 percent.
Can you compress an image without losing any quality?
Only by a sliver. Two routes keep every pixel exactly: removing the metadata, and saving as PNG. We tried both on five photographs.
| Photo | Original | Metadata removed | PNG | JPEG 82 |
|---|---|---|---|---|
| Waterfall in forest (3072 x 2048) | 2,885 KB | 2,862 KB | 3,963 KB | 375 KB |
| Cityscape with signage (3904 x 2602) | 3,145 KB | 3,120 KB | 4,039 KB | 364 KB |
| Sunflower, blurred background (2434 x 1697) | 2,738 KB | 2,738 KB | 3,491 KB | 198 KB |
| Mountain range and sky (2971 x 1615) | 5,467 KB | 5,457 KB | 3,053 KB | 238 KB |
| Waterfall, close crop (1280 x 959) | 369 KB | 369 KB | 2,699 KB | 341 KB |
Removing the metadata saved at most 0.8 percent, and nothing at all on two of them. It is still worth doing for privacy, and Remove EXIF data from photos does it without touching the pixels.
PNG went the other way. Even after shrinking to 1600 pixels, the PNG was larger than the full-size camera file for three of the four camera photos.
So for a photograph, "without losing quality" really means without losing quality you can see. That is a JPEG, WebP or AVIF quality setting, chosen with care.
What quality setting looks the same as the original?
We saved each photo at 1600 pixels at ten quality settings and compared every file with a lossless copy. The first table is file size.
| Photo | Quality 100 | Quality 90 | Quality 82 | Quality 70 | Quality 50 | Quality 30 |
|---|---|---|---|---|---|---|
| Waterfall in forest | 2,033 KB | 522 KB | 375 KB | 277 KB | 199 KB | 139 KB |
| Cityscape with signage | 2,450 KB | 500 KB | 364 KB | 274 KB | 204 KB | 148 KB |
| Sunflower, blurred background | 1,805 KB | 302 KB | 198 KB | 139 KB | 100 KB | 74 KB |
| Mountain range and sky | 1,772 KB | 338 KB | 238 KB | 173 KB | 123 KB | 88 KB |
| Waterfall, close crop | 968 KB | 395 KB | 341 KB | 205 KB | 170 KB | 106 KB |
Quality 100 files were 5 to 9 times larger than quality 82 on the four camera photos. The second table shows what that extra size bought.
| Photo | Quality 100 | Quality 90 | Quality 82 | Quality 70 | Quality 50 | Quality 30 |
|---|---|---|---|---|---|---|
| Waterfall in forest | 0.9992 | 0.9823 | 0.9700 | 0.9548 | 0.9341 | 0.9048 |
| Cityscape with signage | 0.9990 | 0.9885 | 0.9820 | 0.9736 | 0.9608 | 0.9414 |
| Sunflower, blurred background | 0.9981 | 0.9804 | 0.9748 | 0.9693 | 0.9630 | 0.9531 |
| Mountain range and sky | 0.9983 | 0.9846 | 0.9768 | 0.9672 | 0.9538 | 0.9348 |
| Waterfall, close crop | 0.9998 | 0.9957 | 0.9897 | 0.9398 | 0.9207 | 0.8851 |
At quality 82, every photo scored 0.970 or higher. Quality 100 scored above 0.998, so all that extra size bought less than 0.03 of a point.
Below 70 the loss speeds up. On every camera photo, each step down from 82 to 70, then 50, then 30 cost more similarity than the step before it.
So start around 80 to 82, compare the preview with the original, and only then go lower. In the free online image compressor, you can set the quality and see the new file size before you download.
What does the similarity score mean?
The score is SSIM, a standard measure of how alike two pictures are. It compares brightness, contrast and structure in small windows across the whole image. A score of 1 means identical.
It is a number, not an eye. Our version reads brightness only, so a colour shift does not lower it. Look for yourself too. Blocky skies, halos around lettering and smeared skin are the usual first signs of too much compression.
The photo matters as much as the number. At quality 30 the sunflower, mostly soft background, still scored 0.953, while the forest waterfall, all leaves and ripples, fell to 0.905.
Does saving a JPEG again lose quality?
Less than you have been told, if nothing changes between saves. We fed each photo's quality 82 file back in and saved it again at 82 and the same width, ten times over.
| Photo | 70 once | 70, then 82 | 82 once | 82, ten times | 82, then 95 |
|---|---|---|---|---|---|
| Waterfall in forest | 0.9548 | 0.9468 | 0.9700 | 0.9699 | 0.9692 |
| Cityscape with signage | 0.9736 | 0.9686 | 0.9820 | 0.9814 | 0.9815 |
| Sunflower, blurred background | 0.9693 | 0.9663 | 0.9748 | 0.9746 | 0.9746 |
| Mountain range and sky | 0.9672 | 0.9617 | 0.9768 | 0.9768 | 0.9763 |
| Waterfall, close crop | 0.9398 | 0.9373 | 0.9897 | 0.9896 | 0.9877 |
After ten saves the score moved by at most 0.0006, and the file size by at most 0.5 percent. JPEG discards most of what it will discard on the first save.
Changing the setting is what hurts. Saving a quality 70 file again at 82 made it 12 to 13 percent larger than the quality 70 file, and it scored lower, on all five photos.
Saving the quality 82 file again at 95 made it 38 to 46 percent larger, and again it scored lower on every photo. A higher number cannot bring back detail that is gone. It only stores the damage more carefully.
We did not test cropping or resizing between saves, which changes the pixels each time. Keep your original, and make every new version from it rather than from the last copy.
Should you save at quality 100?
Not from a JPEG. Saving each full-size camera photo at quality 100 made all five files larger than the file they came from, some nearly three times larger, with nothing gained.
One of our photos had already been saved at quality 87 before we got it. Saved again at 87, it came out within 0.2 percent of its original size and scored 0.99999.
Saved at 90 instead, the same photo was 7 percent larger than the original and scored lower. Going above a photo's own quality costs bytes and adds no detail.
Is WebP better at the same file size?
Yes, in our test. Against the JPEG closest to it in file size, WebP at quality 82 scored higher on all five photos.
At the same quality number, WebP made a smaller file that scored a little lower on four of the five, which is why comparing by number misleads.
Our JPG vs PNG vs WebP vs AVIF comparison has the full-size figures, and Convert JPG to WebP switches a photo in your browser.
If the file is still too big at a quality you are happy with, change the width instead. How to reduce image size for a website, measured shows resizing takes off far more than quality does.
How was this measured?
Each photo went through this site's engine in Chrome on 26 September 2026, the same code that makes your download. Widths are the long edge, and the close crop stayed at its own 1280 pixels.
SSIM compares brightness in 8 by 8 pixel windows against the PNG export at the same size, decoded by Chrome itself. The photographs are the five Wikimedia Commons images behind this site's file size pages.
Last updated: