How to Compress JPG Without Losing Quality

The short answer

Save at JPEG quality 70-80. For most photos, that's the sweet spot where the file is dramatically smaller (often 60-80% less than the source) and no human looking at the image at normal viewing distance can tell.

Anything above 90 barely shrinks the file. Anything below 50 starts to show JPEG artifacts — blocky 8x8 grids and color banding in smooth regions like skies and skin.

The reason 70-80 works as a universal default is that JPEG's quantisation tables are tuned to discard frequencies in roughly that order. Each step down from 100 discards a wider band of high-frequency information, but the perceptual cost stays close to zero until you cross into the 60s. Below that, the human visual system starts noticing the missing detail because the discarded frequencies overlap with the ones we actively use for edge detection.

If you only remember one number, make it 75. It works for portraits, landscapes, product shots, and screenshots-of-photos. The exceptions are narrow: extreme close-ups of skin where you want every pore, or images destined for further editing (in which case you shouldn't be using JPEG at all).

Why JPEG compression works

JPEG splits the image into 8x8 pixel blocks and stores each one in the frequency domain. The high-frequency parts (sharp edges, fine texture) get quantised away first because the human eye is less sensitive to them.

That's the trick: throw away what nobody will miss. Done well, you save 70% of the file size. Done badly (quality < 30), the throw-aways pile up and the image looks crunchy.

The transform under the hood is the discrete cosine transform (DCT). Each 8x8 block produces 64 coefficients, and the lower-frequency coefficients (top-left of the block) carry the bulk of the visual weight. The quality setting controls how aggressively the high-frequency coefficients get divided down before being stored as integers. At quality 100, the divisor is 1 for most coefficients. At quality 50, common divisors are 16 or more, which is why fine texture vanishes first.

Chroma subsampling adds another layer. Most JPEG encoders default to 4:2:0, which stores color resolution at half the luminance resolution in both axes. Your eyes have far fewer color-sensitive cones than brightness-sensitive rods, so this is essentially free savings. It's also why pure red text on a pure green background renders poorly in JPEG — colored edges land right where the encoder cuts corners.

Common mistakes

Re-saving a JPG repeatedly. Every save loses a little more — open, edit, save once. If you'll edit again, convert to PNG first.

Compressing twice. Re-compressing a JPG with the same tool at quality 70 doesn't make it 30% smaller again; the gains are tiny and the loss compounds.

Compressing before resizing. If you'll display the photo at 800px wide, resize to 800px first, then compress. Compressing a 4000px source and then resizing wastes effort.

Trusting the file extension. A file called photo.jpg that came out of a phone editing app might already have been recompressed two or three times before it reached you. Check the file size against the resolution — anything under about 0.2 bytes per pixel for a photo has been pushed hard already, and another pass will visibly degrade it.

Using one quality setting for every image type. Photos tolerate quality 70 well. Screenshots of UI panels, charts, or anything with sharp text and flat regions need quality 85+ or, better, a switch to PNG or WebP-lossless. JPEG ringing around text is the most common giveaway of over-compression.

Step-by-step with freeimgtool

The whole process on our free compressor:

• Open the Image Compressor.
• Drop a JPG or click to browse.
• Move the quality slider to 75.
• Click Compress.
• Check the Before / After / Saved row.
• Click Download.

Real-world numbers

A typical 12-megapixel phone photo (4032 x 3024) straight from the camera lands in the 3-5 MB range. At quality 80, expect output in the 600-900 KB range with no visible difference at any normal viewing distance. At quality 70, the same photo usually compresses to 350-550 KB, still clean for web and social use.

A 1920 x 1080 landscape exported from Lightroom at quality 100 often runs 1.5-2.5 MB. The same file at quality 80 is typically 250-450 KB — a 6-8x reduction with no perceptual loss in side-by-side blind tests. The savings come entirely from the high-frequency noise that the camera sensor and demosaic step inject; that noise was never visible information.

When you scale down a 4000 px source to 1600 px before compressing, the savings stack. The same photo that landed at 700 KB after a quality-80 pass at full resolution drops to about 200 KB at 1600 px, quality 80. Resize first, then compress — the order matters because the resize removes the high-frequency content that the JPEG quantiser would have discarded anyway, but without the encoding overhead.

When NOT to compress at all

Master files. Anything you'll come back to and re-edit — keep a lossless original (RAW, TIFF, or PSD). Compress the export, not the master.

Medical imaging, scientific photography, evidence photos. Anything where a courtroom or lab might need to verify the original pixels. Lossy compression destroys that chain of trust because there's no way to prove what was lost.

Print at large sizes. A 24-inch print at 300 DPI needs roughly 7200 px on the long edge. JPEG compression artifacts that hide on a phone screen become visible across that area. Use quality 95+ or PNG/TIFF for print output.

Images that have already been compressed hard. Re-encoding a quality-50 JPEG at quality 80 doesn't restore detail; it just adds more loss on top of the existing loss. If you receive a heavily compressed file and need a smaller version, accept the existing artifacts and don't compress further.

Frequently asked questions