Compress PDF

A PDF that started life as a Word document or a design export is fundamentally a container of text, vector shapes and embedded fonts, and shrinking it means tightening that structure — removing duplicate font subsets, recompressing any embedded raster images that are already present, stripping unused metadata and unreferenced objects. None of this touches the text itself, which stays exactly as selectable and searchable as before. The saving here is typically modest, often in the range of a few percent to perhaps twenty, because a well-formed document does not carry that much slack to begin with.

A PDF that started life as a scanner, by contrast, is really a sequence of full-page photographs wrapped in PDF packaging, with no real text at all — what looks like text is pixels arranged to resemble it. Shrinking that kind of file means re-encoding each page image at a lower resolution or quality, exactly the same trade-off as compressing a JPEG. The saving here is dramatic, routinely 70–90%, precisely because a full-page scan at print resolution carries far more pixel data than a screen actually needs to display it.

Treating both cases as one "compression level" would hide the decision that actually matters: one path is safe to apply to anything, the other permanently converts real text into an image and cannot be undone. Keeping them as two explicit modes means the choice is visible rather than buried in a slider.

The reliable test is whether text in the document can be selected and copied. Open the file, try to click and drag across a line of text, and if a text cursor and a selection highlight appear, the document has real text and lossless mode is what you want — rasterising it would make it larger, not smaller, since re-encoding crisp vector text as a pixel image is a poor trade in both directions. If clicking produces no selection and the whole page behaves as a single image, it is a scan, and rasterise mode is the appropriate one.

A mixed document — a contract with a scanned signature page appended to otherwise digital pages — will test as having real text overall, and lossless mode is the right choice for it; the embedded scan image within it still benefits from whatever recompression lossless mode is able to apply to embedded images specifically, just without touching the rest of the document's structure.

The size difference you sometimes see between this tool and a server-side PDF compressor is not a limitation to be embarrassed about — it is the direct consequence of processing entirely inside the browser rather than sending the file to a server with a full PDF engine and dedicated image codecs. A server-side tool can subset fonts down to only the glyphs actually used, re-encode embedded images with a purpose-built encoder tuned specifically for that job, and rebuild the entire object graph from scratch. Replicating all of that inside a browser tab, without a backend, is a materially different engineering problem, and this tool is honest about doing a smaller, safer subset of it rather than claiming parity it cannot deliver.

What you get in exchange is straightforward: the document never leaves your device, so a compressor pass on a contract or a medical record never becomes a data-handling question, and there is no page count limit, no monthly quota and no paywall gating the difference between "small enough to compress for free" and "pay to compress the rest."