Compress a PDF file

Shrink a PDF locally using modern object streams, image re-encoding, and optional document-metadata cleanup. Pages are never rasterised: text stays selectable, vector graphics remain sharp, and pages retain their original dimensions. Embedded photographs are re-encoded as JPEG, which is where most of the saving comes from and which does discard some detail.

Three levels decide how much detail to trade. Light and balanced re-encode photographs at their original pixel dimensions, so every pixel survives and only the encoding quality changes. Strong also halves the resolution, which is what lets it reach a genuinely small file rather than a slightly smaller one. Balanced is the default.

Compression is not guaranteed for every document. A PDF of nothing but text has little to reclaim, and pictures stored in a form the tool does not read safely, such as palette colour, are left exactly as they are. In those cases the output stays close to the source size and the tool says so rather than claiming a misleading reduction. An image is only replaced when the new version is genuinely smaller.

Processing happens in a dedicated worker and the document never leaves your browser. Enable metadata removal to clear common title, author, subject, and keyword fields before saving. Password-protected PDFs must first be unlocked by their owner. Device memory is the practical limit for large documents.

PDF size comes from two places: the pictures and the packaging. Scanned pages carry large photographs, and office applications leave duplicate or loosely packed objects behind. This tool addresses both, re-encoding the photographs it can read and rewriting the document with efficient object streams.

Use this one when selectable text and vector diagrams must survive and some loss of photographic detail is acceptable. When the original image data must remain untouched, do not use it. It cannot promise a portal-specific figure either; a strict target needs the target tool, which verifies the final bytes.

After processing, open the downloaded copy and check several image-heavy pages, the searchable text, and any form fields that matter. Keep the original until the result has been accepted by whatever it is going to. If the output is larger, the tool says so instead of pretending every PDF contains removable waste.

Metadata removal is useful before publishing a document, but it is not a complete redaction system. Visible names, comments flattened into pages, attachments, form values, and text beneath black rectangles can still disclose information. Remove sensitive content in an appropriate editor and inspect the exported PDF. FileGizmo’s option only targets common document-information fields and makes no broader anonymity promise.

Because the operation runs locally, confidential reports do not travel to a compression service. Browser memory and processor speed determine practical limits. Close memory-heavy tabs for large documents, use a current desktop browser when possible, and avoid interrupting the page until the result is ready.

Use the downloaded filename to keep the optimized copy associated with its source, then record which version was submitted or published. A smaller byte count is worthwhile only when the document still opens correctly in the readers, archival systems, or portals that matter. Local processing protects the transfer path; careful review protects the final outcome.