PDFree › Compress PDF › What gets removed
Plain-English explanations · Updated June 2026
When PDFree compresses your PDF, the compression report lists exactly what was found and removed. This page explains what each item is, why it exists in the PDF, and how much space removing it typically saves.
The numbers in your compression report are real. PDFree analyses your specific file — "3 thumbnails removed" means exactly 3 thumbnail objects were found and removed from your PDF. The report is not an estimate or a template. Size figures on this page are labeled as typical examples to give context, since every PDF is different.
What it is: XMP (Extensible Metadata Platform) is an XML document embedded inside the PDF that stores document properties — author, title, creation date, the software used to create the file, and copyright strings.
Why it exists: Adobe created XMP as a standard way for applications to embed rich metadata. InDesign, Acrobat, Illustrator, Word, and most PDF generators add it automatically when saving. A typical XMP stream is 5–50 KB of XML text — invisible in the document but present in the file.
Space saved: Varies by document. Typical example: 5–50 KB. Small on its own, but meaningful when combined with other cleanup. PDFree removes XMP streams on Standard and Maximum presets; Light leaves them intact for maximum compatibility.
What they are: Small preview images of each page, stored inside the PDF itself. They look like the page panel in Adobe Acrobat — miniature versions of your pages.
Why they exist: Adobe Acrobat generates and saves thumbnails to speed up its own page panel rendering. When you scroll through pages in Acrobat, it displays these pre-rendered images instead of rendering pages on the fly. Other PDF viewers (Chrome, Firefox, Preview, most mobile apps) render their own thumbnails and never read these embedded ones.
Space saved: Varies by document. Typical example: 2–10 KB per page. A 100-page scanned document might carry several hundred KB of thumbnail data — invisible overhead. PDFree removes thumbnails on all presets including Light, since no viewer needs them from the file.
What it is: A private dictionary added by Adobe applications (InDesign, Illustrator, Photoshop) that stores application-specific edit session data — placed image paths, layer information, or notes used internally by the software.
Why it exists: It allows Adobe applications to "remember" the last edit state of the document so they can re-open it and continue editing with full fidelity. Once the PDF is finalized for distribution, this data serves no purpose for anyone reading the document.
Space saved: Varies significantly by source application. Typical example: 5–200 KB. Documents produced by InDesign or Illustrator tend to carry the largest PieceInfo blocks. PDFree removes PieceInfo on Standard and Maximum presets.
What they are: The PDF Info dictionary — a simpler, older form of metadata storing title, author, subject, keywords, creator, and producer strings. Every PDF has one.
Why they exist: The PDF specification includes a standard metadata dictionary so applications can tag documents with basic properties. Word, LibreOffice, Google Docs, and every PDF printer automatically populate these fields.
Space saved: Minimal by itself. PDFree clears the text content but keeps the structure. The real value is reducing document fingerprinting — author names, software versions, and creation timestamps are cleared from the output.
What they are: Identical image data stored as multiple separate objects in the PDF, even though the visual output is the same image appearing in multiple places.
Why they exist: When PDFs are assembled from multiple sources — merged documents, combined chapter files, or repeatedly pasted images — the same logo, watermark, or diagram can be independently embedded each time it appears. A 200-page report with a company logo on every page might contain 200 separate copies of the same 50 KB JPEG instead of one shared reference.
Space saved: Proportional to image size × (number of duplicates − 1). Example scenario: a 50 KB logo embedded 10 times = 450 KB of redundant data. The actual amount depends on how many duplicates exist and how large they are. PDFree identifies duplicates by content fingerprint and rewires all references to a single stored copy.
What it does: JPEG images inside PDFs often have higher quality settings than necessary for the intended use. PDFree recompresses them at a lower JPEG quality (82% for Standard, 72% for Maximum) and optionally downsamples to a lower DPI (150 for Web, 96 for Email).
What gets skipped: Images with CMYK color profiles (would shift colors), images with transparency (JPEG doesn't support alpha), 1-bit black-and-white images, and images where recompression would save less than 10% — all are left untouched.
Space saved: For image-heavy PDFs this is typically the largest single source of savings. Example scenario: a scanned document with 20 MB of 300 DPI JPEG images recompressed to 4–6 MB at 150 DPI. Actual results depend on the number of images, their original quality, and the chosen preset. Text-only PDFs see no benefit here since they contain no raster images.
What it does: PDF content streams (which describe page layout, text positions, drawing instructions) and font data are compressed using the DEFLATE algorithm. Many PDF generators use the default compression level (level 6) or even a low level (level 1) for speed during export.
How PDFree improves this: PDFree uses the browser's native CompressionStream API to re-deflate these streams at a higher level. The data is first decompressed, then recompressed at the optimal level. Only streams that compress at least 5% more are replaced — the rest are left as-is.
Space saved: Depends on how suboptimally the original was compressed. Typical example: 3–15% of non-image stream data. Documents with many pages, complex layouts, or large embedded fonts benefit most. No content is changed — only the compression efficiency.
What it does: Traditional PDFs store each object (font, image, page description, metadata) individually with a cross-reference table listing each object's byte position. PDF 1.5 introduced object streams, which pack multiple objects together and compress them as a group, replacing the per-object overhead.
Why it matters: Small objects (like metadata dictionaries and page tree nodes) have a high overhead-to-data ratio when stored individually. Grouping them in an object stream and compressing together lets DEFLATE find patterns across objects, typically achieving better compression than individual storage.
Space saved: Varies by document structure. Typical example: 10–30% of cross-reference overhead. Most effective for text-heavy documents with many pages and shared resources. PDFree enables object streams on Standard and Maximum presets; Light skips them for maximum compatibility with older PDF readers.
PDFree explains what changed during compression
After compressing your PDF, the report shows exactly which of these were found and removed — with real numbers from your specific file, not estimates.
Compress a PDF →