Compress PDF files

What this compress PDF tool can actually make smaller

A PDF is a container, not one compression format. A page may contain selectable text, embedded fonts, vector lines, form fields, annotations, and several different kinds of image. Those parts respond to entirely different treatments. Rewriting a sentence that is already stored as compact drawing instructions will not save useful space. Re-encoding a large photograph often will. That is why this tool starts by inspecting the file instead of presenting a theatrical progress bar and hoping the output happens to be smaller.

The current browser-safe path is deliberately narrow: ordinary JPEG image objects stored with PDF's DCTDecode filter, in a plain RGB color space, without masks or a specialized filter chain. Those pictures are decoded and encoded again at the level you choose. Text is not converted into pixels. Page dimensions are not reduced. Fonts, links, vector marks, and form structures are not selected as compression targets. When the file contains only those things, the honest result is that this tool has nothing to compress.

PDF, in brief

PDF

Full name	Portable Document Format
Extension	.pdf
Format type	Fixed-layout document
MIME type	application/pdf

The final byte comparison is the real result

Making an individual JPEG smaller is not enough. The PDF writer also has to serialize the document's catalog, page tree, cross-reference information, dictionaries, and streams. That structural rewrite can add bytes. A picture might lose 20 KB while the rebuilt container gains 25 KB, leaving a candidate larger than the source even though the image step worked. Tools that report only the image saving can call that a success. This one compares the complete candidate file with the complete original.

A download appears only when two independent facts are true: at least one supported JPEG was actually replaced by a smaller encoding, and the finished PDF has fewer bytes than the file you handed in. If either test fails, the candidate is discarded. You get a receipt saying that no smaller copy was created and why. There is no renamed original presented as a win, no zero-percent "compressed" download, and no larger file hidden behind an optimistic percentage. The source File object remains the source for every retry.

Both compression levels lose image detail

JPEG recompression is lossy. That statement matters more than a number printed beside a slider. Browser encoders accept a quality hint, but the hint is not a standardized measure of human-visible quality and is not comparable across every browser version. Calling one setting "72% quality" would imply precision that does not exist. The tool therefore offers two choices with consequences rather than invented measurements.

Balanced uses the gentler pass and is the sensible first attempt for a report, application, or document that may be printed. Stronger compression discards more photographic detail for a better chance of a useful byte reduction. Neither changes the pixel dimensions of the embedded image, so the trade is compression artifacts rather than a lower nominal resolution. Inspect faces, fine diagrams, screenshots with tiny text, and smooth gradients in the downloaded copy before replacing anything important.

If the picture exists outside the document, the dedicated image compressor gives you more control because it does not have to preserve a surrounding PDF. You can then rebuild or replace the picture at its source. The file-size guide explains when changing dimensions, cropping, or choosing a different format saves more than another quality pass.

Why some scanned PDFs show no manageable images

"It is a scan" does not tell you how its pages are encoded. One scanner may store every page as a JPEG. Another may use a PNG-like Flate stream, monochrome CCITT fax compression, JBIG2, or JPEG 2000. A scan can also combine a background photograph with separate monochrome text layers. They may look identical in a reader while requiring unrelated decoders and writers. This version handles only the JPEG case that the browser stack can rewrite without flattening the whole page.

Flattening would be an easy way to claim broader support: render each page to a canvas, save a new JPEG, and wrap the pictures in a new PDF. We tested that route. A small text PDF became dramatically larger, and its selectable, searchable, screen-reader-accessible text disappeared. That is not compression. It is destructive rasterization, so this page does not silently fall back to it when it encounters an unsupported stream.

Pages and document behavior still need inspection

The JPEG objects are changed in place within the parsed document rather than copying only visible pages into a new container. That is important because page-only copying can drop document-level structures such as bookmarks and AcroForm data. Our fixtures verify that the output opens in an independent PDF reader, retains its page count and extracted text, and preserves representative links and fillable fields. Those checks are stronger than merely asking the same library that wrote the file to read it again.

They are not a promise about every PDF ever produced. CMYK and ICC-managed JPEGs, soft masks, inline images, chained filters, JPEG 2000, JBIG2, CCITT, malformed object graphs, and unusual encrypted documents remain outside the compression path. Such content is left alone rather than guessed at. Always open the result, move through every page, test important links and fields, and keep the original until the copy has passed the checks that matter to your workflow.

A signed PDF cannot stay cryptographically signed after rewriting

A digital PDF signature covers a defined byte range. Re-encoding one photograph and writing a new cross-reference structure changes those bytes, even when every page looks unchanged. The old signature can no longer prove that the new file is the exact document its signer approved. This tool looks for common signature dictionaries and warns when it finds one, but detection is not a guarantee: signatures and validation data can be represented in several ways, and a damaged file can hide structure from a parser.

Keep the signed original. If your process permits it, compress an unsigned working copy and apply the final signature afterwards. Do not use a smaller download as a replacement for an executed contract merely because the pages look the same. Visual similarity and cryptographic identity are different claims.

Local processing changes the privacy bargain

Most prominent results for PDF compression begin by uploading the document. That may be a reasonable bargain for a public brochure. It is a different decision for a medical letter, tax form, internal report, or signed agreement, especially when the upload happens before the service can tell whether the file contains a compressible image. Here the PDF parser, JPEG decoder, image encoder, writer, and final comparison run in a background worker inside this tab. No file bytes are sent to CreatorValet or an external conversion service.

You do not have to accept that claim on trust. Open the browser's Network panel, clear the log, hand in a PDF, and run the compressor. The local-processing guide shows the exact check. You will see the site's own code load after your first action, but no request carrying your document. Close the tab and the in-memory original, candidate, and temporary object URL are released.