Compress PDF

The weakest level never touches page content. It removes document information — title, author, subject, keywords, producer, creator, creation and modification dates — plus the XMP metadata stream and any thumbnail image embedded in each page. Text and images survive untouched, so the result stays selectable and prints at full quality.

Think of this mode as "drop the extras" rather than "make it small". If the original carried a lot of metadata and thumbnails it shrinks visibly; if it did not, the size barely moves. Re-serialisation can even produce a slightly larger file — when that happens the tool returns the original bytes unchanged, so this mode can never inflate your PDF.

The other two modes work on a fundamentally different principle. Every page is rendered with PDF.js, encoded as a JPEG, and a new PDF is assembled by placing those JPEGs back one per page. Rendering resolution is 150 dpi on Balanced and 96 dpi on Max compression, and JPEG quality is exposed as a slider.

For PDFs whose pages are really images to begin with — anything out of a scanner — this approach is dramatically more effective, because resolution and quality give you two direct handles on the trade-off between legibility and file size.

The moment you pick Balanced or Max, your pages become images. Text becomes something you can look at but not select, copy, or search, and zooming stops resolving beyond the resolution it was rendered at. The UI warns about this. Avoid strong compression on contracts, specifications, and papers — documents you will later need to search or quote.

If the text layer must survive, your options inside a single PDF are limited. Try the text-safe mode first; if that is not enough, the realistic next move is to reduce the number of pages rather than the quality of them.

The text-safe mode removes document information and thumbnails — not the content written on the page. To hide names, addresses, or figures, use markup, not compression. Conversely, while a rasterising mode does destroy the text layer as a side effect, do not treat that as a privacy measure: the information you wanted hidden remains fully visible as an image.

Drop in several PDFs and they are compressed in sequence with the same settings, then handed back as a single ZIP. Shrinking a month of invoices, or bringing a stack of scans down to submission size, becomes one operation. The ZIP itself is also built in the browser.

If the target is an attachment size limit, merging first and compressing the single result is easier to judge, because you are watching the number that actually matters. If one of the inputs is an unusually heavy scan, compressing that one first and merging afterwards keeps the whole operation lighter.