Compress PDF without losing quality

The half that costs nothing

Two of the four methods never touch a pixel. The structural method rebuilds the document into a fresh file and leaves behind orphaned objects and old revisions that edits piled up over time. The font method subsets each embedded TrueType down to the glyphs the pages actually draw, so a font that shipped with 3,000 glyphs stops carrying the 2,700 your document never uses.

Both are lossless in the sense that matters: every page draws the same, every word stays selectable, and every image keeps the pixels it had. On the public run the structural method won 38 of 120 cases and the font method 17. A plain re-save with nothing else changed took the public corpus from 44,524,076 to 41,649,292 bytes, 6.46% off, purely by rebuilding.

The ceiling on this half is low, though. Structural work met the target on only 6 cases and font subsetting on 4. Free savings are still savings, but they are rarely enough on their own to hit a tight number.

The half that costs something

The image method re-encodes placed images at 150 dpi with mozjpeg at quality 75. The raster method renders whole pages at 150 dpi and picks its own quality from a sample of the pages. Both are lossy, and both are where a demand for a small file gets paid for.

The measured range is narrower than people expect. When the raster method won, its median quality was 92 and its maximum was 92. It only reached its floor of quality 60 on 10 of the 120 cases, and that floor exists so the method refuses to keep going into mush. At quality 92 the difference is invisible on a screen. At 60 you can see it on a photograph and not on a printed page of text.

150 dpi is the other lever, and it is a deliberate one. It is above the 96 dpi a screen shows and below the 300 dpi a print shop wants. If you are compressing something that will be printed at full size and inspected closely, that is the number to think about, not the quality figure.

Text quality is not part of the trade

The one thing never traded away is the text. Even the raster method, which turns the visible page into an image, redraws the page's own text on top in invisible render mode, so selection, search, copy and paste and screen readers all keep working on the compressed file.

That is enforced, not promised. Every candidate is read back and compared to the source, and one that lost more than 2% of the non-space characters is refused rather than offered. On the public run, 102 documents carried text, all 102 came back at 100% retention, and zero candidates were refused.

Resolution is the setting people mean

When someone says a compressed PDF looks bad, the cause is usually resolution rather than JPEG quality. An image placed at 600 dpi and re-encoded at 150 dpi has lost three quarters of its pixels, and no quality number brings them back.

150 dpi is chosen to sit above a screen and below a print shop. A page read on a laptop or a phone has nothing to gain above it. A page going to a commercial printer at full size does, and for that job compression is the wrong tool. Send the original.

How to keep quality in practice

Three habits protect quality more than any setting inside the tool.

1. Ask for the largest file the receiving system will take. Every KB you leave on the table is quality you keep. A form that caps at 2 MB should get a 2 MB target, not a 500 KB one chosen out of caution.
2. Cut pages before you cut quality. Removing four pages of appendix photos saves more than any quality setting and costs nothing on the pages that remain.
3. If you have no hard cap at all, the no-target page runs the same engine with the lossless methods doing the work and no number forcing a lossy result.

When you do have a cap, the compress tool takes it directly. The general guide covers which of the four methods your file will lean on, and the compress hub lists the rest of the cluster.