Compare Keep text with Flatten pages: Download the four-page report and add it to Compress PDF. Run Balanced with Keep text, then run again from the original with Flatten pages. Check all four pages and the result note. If flattening produces a smaller file, text becomes page images; if Original kept appears, the downloaded original retains selectable text. This compact report may not shrink; compare actual sizes and the smallest text.
Practice file and tool: Download the synthetic practice file · Open the tool
Where the bytes go
A PDF is a container of objects. Text is stored as characters plus the fonts needed to draw them; a whole book of text fits in a couple of megabytes. Vector drawings — logos, charts, table rules — are stored as drawing instructions and are tiny. Pictures are stored as embedded images, usually JPEG or losslessly compressed pixels, and they are where almost all the size lives. One 300 dpi A4 scan is 2480 × 3508 pixels; stored as a high-quality JPEG that is 1–3 MB per page, and a twenty-page scan lands at 30–50 MB.
Two other things add up quietly. Files that were edited and re-saved several times often carry the old versions of replaced objects inside them, because PDF allows changes to be appended rather than rewritten. And some producers embed full-resolution page thumbnails alongside every page. Neither is visible, and both can be dropped without changing what the document shows.
What compression actually does
One common compression method finds embedded pictures, downsizes them and re-encodes them as JPEG at a lower quality. Text, fonts and vectors are copied through unchanged, which is why a text-only PDF barely shrinks — there is nothing to compress. Toolkitta’s Keep text method does exactly this, on your device, and additionally rewrites the file structure so that leftover objects and page thumbnails are dropped.
The word “lossy” matters. A JPEG re-encoded at a lower quality does not get its detail back, and a downscaled image cannot be upscaled to its original sharpness. That is fine for a document that will be read on a screen, and it is why the original file should be kept: compress a copy for sending, and archive the original.
Choosing a level
The three levels differ in two numbers: the resolution cap for pictures, expressed as dots per inch across the largest page, and the JPEG quality used for re-encoding. The cap is a ceiling — pictures that already have fewer pixels are never enlarged — and any picture whose re-encoded version would not be smaller is left exactly as it was.
- Light: 300 dpi, quality 85. For documents that will be printed or zoomed into: portfolios, technical drawings with photographs, anything with fine print. Savings are modest on files that were already print-ready.
- Balanced: 150 dpi, quality 75. A screen-sharing starting point; inspect fine print before using the output. The default, and the right choice for email attachments, application portals and shared drives.
- Strong: 100 dpi, quality 60. For hard limits such as “under 2 MB”. Scanned text and fine texture may soften; inspect the actual result.
If a portal states a limit, start with Balanced and look at the result row: it shows the before and after file sizes. Check exact bytes in your file manager when a rounded label is ambiguous. If that is still over the limit, try Strong. If Strong is not enough, the document probably has more pages than the limit allows for at any quality — split it, or ask whether the recipient can accept two files.
Compressing to a size limit
When the limit is the whole point — “attachments up to 10 MB”, “upload your CV as a PDF under 2 MB” — pick Target size instead of guessing a level. Click a preset or type the number in KB, and the tool searches seven picture settings, from Light down to a floor of 72 dpi, for the lightest one that lands under the limit. Each pass starts from the original file, so nothing is compressed twice, and the result row reports the resolution that was used alongside the output size.
The search stops at 72 dpi. That limit does not guarantee readable small print; inspect the result before sharing. If a file misses the target even at the floor, the row says so and gives you the smallest version anyway; from there the options are Flatten pages for scans, a larger target if the limit was a guess, or splitting the document into parts that fit.
A worked example
Take a ten-page contract scanned on a phone at 300 dpi and saved as a 28 MB PDF. Each page is a 2480 × 3508 pixel JPEG. At Balanced the cap for an A4 page is about 1754 pixels on the long side, so each page becomes roughly 1240 × 1754 pixels — a quarter of the pixels — and quality 75 stores that in a few hundred kilobytes. The pixel calculation is predictable; the resulting bytes and legibility are not. Measure the actual output and inspect signatures and small print before choosing this setting.
Now take a 40-page report written in a word processor with a logo on each page and three charts. It is 900 KB. The logo is small, the charts are vectors, and there is nothing to downscale, so Balanced returns the original or shaves only the structural leftovers. That is the correct outcome: the tool reports “Original kept” rather than handing back a rewritten file that is no smaller.
When a scan will not shrink
Scanners and print-oriented PDF producers often store pictures in encodings a browser cannot decode directly: JPEG 2000, CMYK or indexed colour, or unusual bit depths. The Keep text method decodes those with the same PDF engine that draws the preview and re-encodes them as RGB JPEGs, so they shrink like any other photo. Black-and-white scans stored as CCITT fax or JBIG2 are left alone because they are already smaller than a JPEG could be, and so are pictures in spot colours. If a file still barely shrinks, switch the method to Flatten pages. For supported PDFs, a rewritten result renders each page at the level’s resolution and stores it as one JPEG picture. Signed documents and XFA forms are still refused; damaged or unusually complex files can fail.
When a flattened rewrite is returned, text is no longer selectable or searchable, links and form fields disappear, and the page is only as sharp as the chosen resolution. If the result says Original kept because the rewrite was larger, the download is still the original with its structures intact. A scan rarely has selectable text to lose, so flattening a scan is a fair trade. Flattening a report or a form is not; use Keep text for those and accept that they may already be as small as they can be.
What is not changed, and what to check
- Open the compressed file and page through it. Look at any photograph and at small print in scans; if either is too soft, go back a level.
- Check that the page count matches, and that forms, bookmarks and links still work if the document has them — Keep text preserves them, Flatten pages does not.
- Remember that compression is not sanitization. Document metadata, hidden annotations and the text under a black box drawn over it are still in the file. Use a proper redaction tool for confidential content.
- Keep the original. The compressed file is for sending; the original is for the archive.
Password-protected PDFs ask for their password on the file card; restricted PDFs open directly. Compressed results are saved without encryption. XFA forms and digitally signed documents are rejected; keep the signed original and use an unsigned copy. This quick tool also refuses empty digital-signature fields; ask for a standard copy without those fields if an unsigned form is rejected.