Which JPEG quality setting? 60%, 75% and 90% measured

We compressed six photos with Image Compressor at 60%, 75% and 90% and measured file size and SSIM. What each setting saves and when to pick it.

Published: 2026-11-12 · Updated: 2026-11-12 · SwiftlyKit

What we tested and how

We ran six camera photos through Image Compressor at three quality settings: 60%, 75% (the default) and 90%. The option to also reduce the dimensions was off, which is the default, so every output kept its original pixel size.

Everything ran on 2026-11-12 in headless Chromium 158 on one desktop PC (AMD Ryzen 7 9800X3D, Windows 11). We called the same worker code the site runs in your browser, so the output bytes are the ones the tool would give you in Chromium. Each setting ran three times per photo.

Quality is measured with SSIM, a standard similarity score, computed on brightness (luma) at full resolution. 1.0 means identical to the source; lower means more change. The reference is the source exactly as the encoder received it, after Chromium decoded it and converted its colours to sRGB. Sizes use 1 KB = 1,000 bytes and 1 MB = 1,000,000 bytes.

  • Landscape: mountains at sunset with a large, smooth sky, 5184 × 3456 px, 4.15 MB, Adobe RGB colour profile.
  • Street: Amsterdam brick facades with fine detail, 4592 × 3448 px, 9.53 MB.
  • Portrait: close-up of a person, 4272 × 2848 px, 7.78 MB.
  • Night: a Budapest bridge at night, ISO 125 long exposure with little noise, 3580 × 2000 px, 5.28 MB.
  • Product: a black smartwatch shot on a phone, 3024 × 4032 px, 2.98 MB, Display P3 colour profile.
  • High-ISO: a starry sky at ISO 3200 with visible sensor noise, 4000 × 2670 px, 13.4 MB.

Results at 60%, 75% and 90%

Each cell shows the output size, the share of the original saved and the SSIM score. Every photo got smaller at every setting.

PhotoOriginal60%75% (default)90%
Landscape4.15 MB314 KB · −92% · 0.958627 KB · −85% · 0.9661.79 MB · −57% · 0.979
Street9.53 MB1.83 MB · −81% · 0.9672.38 MB · −75% · 0.9763.98 MB · −58% · 0.988
Portrait7.78 MB346 KB · −96% · 0.978591 KB · −92% · 0.9851.10 MB · −86% · 0.992
Night5.28 MB697 KB · −87% · 0.954926 KB · −82% · 0.9661.61 MB · −70% · 0.982
Product2.98 MB499 KB · −83% · 0.912871 KB · −71% · 0.9292.20 MB · −26% · 0.966
High-ISO13.4 MB1.11 MB · −92% · 0.8041.64 MB · −88% · 0.8543.35 MB · −75% · 0.943

From 90% to 75%: large savings, small score changes

Moving from 90% to 75% removed a large share of the bytes while SSIM moved only slightly. The landscape dropped from 1.79 MB to 627 KB as its score went from 0.979 to 0.966. The portrait halved from 1.10 MB to 591 KB (0.992 to 0.985). The street photo went from 3.98 MB to 2.38 MB (0.988 to 0.976). Going further to 60% saved less in absolute terms; the street photo only fell from 2.38 MB to 1.83 MB.

The saving also depends on how the original was saved. Camera files stored at a high quality shrink a lot even at 90%: the portrait lost 86%. The phone-shot product photo was already compact and saved only 26% at 90%.

A setting does not guarantee a quality level either. At 75%, SSIM ranged from 0.929 for the product photo to 0.985 for the portrait, and fell to 0.854 for the noisy high-ISO photo.

What the quality slider actually does

With the dimensions unchanged, Image Compressor re-encodes the photo with Chromium's own JPEG encoder. Its output matched PNG ↔ JPG Converter's JPEG at the same quality byte for byte; the landscape at 60% was 313,927 bytes from both tools.

Below 100%, Chromium's JPEG encoder stores colour at half resolution in each direction (4:2:0 chroma subsampling). Brightness keeps full detail, but fine red or blue patterns and coloured text edges lose some sharpness at every setting from 60% to 90%.

When do differences become visible?

We did not run a viewing study, so we cannot name the setting at which people start to notice. SSIM is an objective proxy, and it can mislead in both directions: on flat skies it penalises lost grain that viewers rarely see, while on sharp edges a high score can hide faint ringing. Check the result yourself at 100% zoom, on the screen or print where it will be used. Problems usually show first in these places:

  • Smooth gradients such as skies, where blocks or banding appear.
  • Fine repeating texture such as brickwork, fabric or foliage.
  • Faces and hair, where smoothing looks waxy.
  • High-contrast edges and small text, where halos appear.
  • Dark, noisy areas such as night skies.

High-ISO photos: low scores, a different trade-off

The ISO 3200 night-sky photo scored far lower than the others: 0.804 at 60%, 0.854 at 75% and 0.943 at 90%, against 0.91 to 0.99 for the rest. JPEG compression smooths away much of the random sensor noise, and SSIM counts every removed speck as a difference. Sometimes that makes a photo look cleaner; sometimes faint stars or fine detail disappear along with the noise. For photos like this, compare the settings side by side and consider 90% if the fine points matter.

What we recommend

Image Compressor took 0.11 to 0.23 s per photo on our fast desktop; phones will be slower. Based on these results:

  • Start with 75%, the default. It saved 71 to 92% on our photos.
  • Use 60% for thumbnails, previews and chat images after a check at 100% zoom.
  • Use 90% for prints, product photos people will zoom into and noisy low-light shots.
  • Resize as well when the image will be shown smaller. In our format test, the street photo at JPEG 75% was 423 KB at 1600 px instead of 2.38 MB at full size. Use Also reduce the dimensions in Image Compressor, or Image Resizer; their resizing may give slightly different sizes from our test.
  • Need smaller files at the same quality? In our JPG vs PNG vs WebP test, WebP Converter needed 7 to 31% fewer bytes than JPEG for the same SSIM at 1600 px.
  • Keep the original. The output drops EXIF data such as camera details, date and location, and the colour profile; wide-gamut photos like our Adobe RGB landscape and Display P3 product shot are converted to sRGB.

Limits of this test

Six photos cannot cover every kind of image, and results depend heavily on content and on how the originals were saved. All runs used Chromium 158 on one machine; Firefox and Safari use different JPEG encoders and will give different sizes. SSIM was computed against the colour-converted source, so it measures compression loss, not the conversion to sRGB.

Sample files and credits

The six photos are public domain (CC0) on Wikimedia Commons, and each download was checked against the checksum Commons publishes. CC0 does not require attribution, but we credit the photographers anyway; the file pages are listed under Sources.

Try the related SwiftlyKit tools

Image Compressor

Image Resizer

WebP Converter

PNG ↔ JPG Converter

Sources and further reading

Landscape photo: “Landscape-mountains-sky-sunset (24300570116).jpg” by www.Pixel.la Free Stock Photos, CC0, Wikimedia Commons

Street photo: “Brick house facades in old Amsterdam city” by Fons Heijnsbroek, CC0, Wikimedia Commons

Portrait photo: “Young woman in leather jacket (Unsplash).jpg” by Léa Dubedout (leadbt), CC0, Wikimedia Commons

Night photo: “Széchenyi Chain Bridge in Budapest at night.jpg” by Wilfredor, CC0, Wikimedia Commons

Product photo: “Withings ScanWatch 2 black.jpg” by Susie Felber, CC0, Wikimedia Commons

High-ISO photo: “Stars in the night sky (Unsplash).jpg” by Shazia Mirza (shazia23), CC0, Wikimedia Commons

Wang et al.: The SSIM index for image quality assessment

MDN: serializing a canvas image