The same pages, up to twelve times lighter

A PDF that weighs more than it needs to costs something every time: when it is stored, when it is emailed, when it is downloaded on a phone, when it is still kept decades from now. The team measured the documents on this site before and after the engine learned to leave out what nothing reads, and not one of them lost a single mark.How the figures on this page were measured

The documents, before

5 663.7 kB

The same documents, now

872.9 kB

Saved

84.6 %

What did not change

The saving comes from what a document carries rather than from what it shows: an embedded font now describes only the characters the document actually draws, instead of every character the font knows how to draw. A three-page letter no longer carries an alphabet it never uses.

The documents measured, before and after

The documents measured, before and after
Document Before Now Saved
Cutting out what you draw 107.2 8.9 91.7 %
Text and shapes filled with a gradient 108.7 10.6 90.3 %
Shrunk headings and a turned label 108.6 10.6 90.2 %
Breaking a paragraph up, and aligning it 109.0 10.9 90.0 %
Coloured cell borders and fills 109.1 10.9 90.0 %
An invoice laid on a letterhead template 218.9 23.1 89.4 %
Text poured into columns 109.6 11.6 89.4 %
A cell's margin, border, inner room and fill 109.7 11.7 89.4 %
Paragraph indents and spacing 109.8 11.8 89.2 %
Break points after a slash or a hyphen 109.8 11.9 89.2 %
Soft hyphens and no-break spaces 109.9 12.0 89.0 %
Stamps fitted to a rectangle 110.1 12.3 88.9 %
Grey, screen and printing colours 110.4 12.5 88.7 %
Changing style in the middle of a paragraph 110.6 12.7 88.5 %
Links and marks on placed text 110.7 12.8 88.4 %
Lists made from an indent and a tab stop 110.9 13.1 88.2 %
One pattern, repeated across whatever it fills 111.4 13.4 88.0 %
One drawing, placed many times 111.1 13.4 87.9 %
Breaking an overlong word 111.3 13.6 87.8 %
Fonts embedded in the file 111.3 13.7 87.7 %
A self-paginating table 223.3 27.5 87.7 %
The same numbers, said so they mean something 111.7 13.8 87.6 %
Columns measured from their cells 111.9 14.1 87.4 %
A contents page with dot leaders 111.9 14.2 87.3 %
Tabs and tab stops 112.4 14.7 87.0 %
Cells merged down the rows 112.6 15.1 86.6 %
Overlong text in a narrow cell 112.9 15.3 86.4 %
A sheet that measures what it says 113.1 15.5 86.3 %
Colour the document defines 113.4 15.5 86.3 %
Custom page labels 113.5 15.8 86.1 %
A colour chart the drawing points into 113.8 16.0 86.0 %
One hatch, laid down in any colour you like 113.7 16.0 86.0 %
One colour, seen through four stencils 114.2 16.4 85.6 %
One rule instead of a hundred colours 114.4 16.8 85.3 %
A multi-page invoice table 115.0 17.2 85.0 %
Several plates at once, each at its own strength 115.0 17.4 84.9 %
A picture fitted into a rectangle 116.0 17.9 84.6 %
A sheet as it goes to the press 115.2 17.9 84.4 %
A handbook that says how it wants to be read 116.7 19.2 83.5 %
The device a file's colours are meant for 119.2 21.7 81.8 %
An archivable PDF/A-3 form 120.8 22.8 81.1 %
Factur-X: the electronic invoice 120.9 23.0 80.9 %
A file that explains its own entries 122.2 24.6 79.8 %
A register whose pages say where they came from 124.0 26.4 78.7 %
Pictures inside table cells 172.7 75.1 56.5 %
Images on a page 9.3 9.3 0.0 %
All the documents 5 663.7 872.9 84.6 %

One document at the foot of the table did not move at all, and could not have: it carries no font. It is made of pictures alone and draws not a single letter, so there was no alphabet in it to leave out.

The same document, in each language

Every example is written twice, once per language, and the words decide how much of a page there is. So the same document does not weigh the same in English as in French. The table below reads the size off each file this site serves, in both languages at once, so it says the same thing whichever language you are reading it in.

Document English French Difference
The watermark a trial copy carries 63.4 66.8 3.3
A page whose letters cost one byte each 25.8 24.3 1.5
Gradients in a line, in a circle, and read from a function 12.8 11.6 1.3
One rule instead of a hundred colours 16.8 15.7 1.1
A document signed with a demonstration certificate 19.4 20.5 1.0
Links and marks on placed text 12.8 11.8 1.0
An archivable PDF/A-3 form 22.8 21.8 1.0
One colour, seen through four stencils 16.4 15.5 0.9
A page that carries no font 6.4 7.2 0.9
A list whose pictures speak too 71.7 72.5 0.8
A delivery note 70.5 71.3 0.8
Coloured cell borders and fills 10.9 10.2 0.8
One pattern, repeated across whatever it fills 13.4 12.6 0.8
Showing page one before the file has arrived 1 207.1 1 207.8 0.7
Transparency and blend modes 12.8 12.1 0.7
A till receipt on a paper roll 17.3 16.6 0.7
A slideshow that runs itself 22.8 23.5 0.7
A boarding pass 17.8 18.4 0.7
Colour the document defines 15.5 14.9 0.7
A register whose pages say where they came from 26.4 27.0 0.6
Words that follow a curve 14.9 15.5 0.6
Breaking an overlong word 13.6 14.2 0.6
Shrunk headings and a turned label 10.6 11.2 0.6
A handbook that says how it wants to be read 19.2 19.9 0.6
Badges of one size, names of every length 16.5 17.1 0.6
A tender pack of three lots 16.4 17.0 0.6
Cells merged down the rows 15.1 15.7 0.6
Making the file smaller 13.3 12.7 0.6
A gallery of picture formats 87.8 88.4 0.6
Soft hyphens and no-break spaces 12.0 11.5 0.6
Text poured into columns 11.6 11.1 0.5
PDF417 codes 16.3 16.8 0.5
A report that lays itself out 15.6 15.1 0.5
A sheet as it goes to the press 17.9 17.4 0.5
A plan that shows what can be read 15.0 15.5 0.5
QR codes 17.0 16.5 0.5
A two-page services contract 76.9 77.4 0.5
A self-paginating table 27.5 27.0 0.5
Custom page labels 15.8 15.3 0.5
A commercial invoice 72.4 72.9 0.5
Pictures inside table cells 75.1 74.7 0.5
A figure that says which headings reach it 60.5 61.0 0.5
A document that reads freely and asks not to be copied 13.4 13.9 0.5
Changing style in the middle of a paragraph 12.7 12.2 0.4
A font behind the font 27.4 27.9 0.4
A sheet prepared for the printing works 19.9 19.5 0.4
A statement released in part 18.4 18.0 0.4
Overlong text in a narrow cell 15.3 14.9 0.4
A page set in a font from 1990 23.9 24.3 0.4
Paragraph indents and spacing 11.8 11.5 0.3
A sheet of shelf labels 19.9 20.2 0.3
One hatch, laid down in any colour you like 16.0 15.6 0.3
A file that explains its own entries 24.6 24.3 0.3
A document that can be read aloud 20.5 20.2 0.3
Tabs and tab stops 14.7 14.3 0.3
Breaking a paragraph up, and aligning it 10.9 11.2 0.3
A colour chart the drawing points into 16.0 15.7 0.3
A payslip 72.0 72.4 0.3
One drawing, and each reader sees their own 15.1 15.4 0.3
A cell's margin, border, inner room and fill 11.7 11.4 0.3
Fonts embedded in the file 13.7 13.4 0.3
Code 128 and retail barcodes 14.5 14.7 0.3
A contents page with dot leaders 14.2 13.9 0.3
A delivery note that knows its own order 16.9 16.7 0.3
The notes a program leaves itself 12.0 12.2 0.3
Stamps fitted to a rectangle 12.3 12.0 0.3
Three statements in one file 20.7 20.9 0.3
A month's report, its charts drawn in curves 14.8 14.5 0.3
Lists made from an indent and a tab stop 13.1 12.9 0.2
Two orders, one frame, one bottom edge 15.4 15.2 0.2
A story that carries on further in 15.7 15.5 0.2
A document that asks for a password before it opens 13.3 13.5 0.2
Columns measured from their cells 14.1 13.9 0.2
Aztec codes 15.6 15.8 0.2
Dashed and dotted lines 14.7 14.9 0.2
A picture fitted into a rectangle 17.9 17.7 0.2
A page and a stamp inside table cells 33.0 32.8 0.2
One letter each, in one file 21.4 21.6 0.2
The same numbers, said so they mean something 13.8 14.0 0.2
A multi-page invoice table 17.2 17.0 0.2
Factur-X: the electronic invoice 23.0 23.2 0.2
An agreement received and added to without a byte moved 26.8 26.6 0.2
An interactive form 16.9 17.0 0.1
One drawing, placed many times 13.4 13.3 0.1
A snag list, with every defect marked on the plan 34.4 34.2 0.1
Letters you draw yourself 14.2 14.1 0.1
A landscape certificate 68.9 68.8 0.1
Cutting out what you draw 8.9 9.1 0.1
Break points after a slash or a hyphen 11.9 11.8 0.1
The device a file's colours are meant for 21.7 21.6 0.1
A sheet that measures what it says 15.5 15.6 0.1
Several plates at once, each at its own strength 17.4 17.3 0.1
Text and shapes filled with a gradient 10.6 10.6 0.0
Prices crossed out, a shadow behind the title 15.1 15.1 0.0
Four kinds of mesh, every corner its own colour 14.1 14.2 0.0
Data Matrix codes 16.0 16.1 0.0
Inks on plates of their own 18.5 18.5 0.0
An invoice that asks before it claims 23.6 23.7 0.0
An invoice laid on a letterhead template 23.1 23.2 0.0
Grey, screen and printing colours 12.5 12.5 0.0
Internal and external links 14.1 14.1 0.0
What a font knows about its own letters 14.6 14.6 0.0
An agreement built on three supplied templates 61.8 61.8 0.0
Detail

A kilobyte on this site is a thousand bytes. The table below counts each document to the byte.

All the documents together: 5 663 690 bytes before, 872 872 bytes now.

Both figures on a row are measured on the English document, so that the two of them come off one and the same file: the same pages written in French make another document, which is not the one that was weighed before. The English figure can be checked by downloading the file behind the row, and the second table below says what each document weighs in each language.

Each document, measured in bytes
Document Bytes before Bytes now
Cutting out what you draw 107 155 8 946
Text and shapes filled with a gradient 108 742 10 591
Shrunk headings and a turned label 108 630 10 626
Breaking a paragraph up, and aligning it 108 987 10 869
Coloured cell borders and fills 109 060 10 935
An invoice laid on a letterhead template 218 913 23 133
Text poured into columns 109 635 11 602
A cell's margin, border, inner room and fill 109 685 11 673
Paragraph indents and spacing 109 754 11 804
Break points after a slash or a hyphen 109 804 11 880
Soft hyphens and no-break spaces 109 923 12 037
Stamps fitted to a rectangle 110 127 12 267
Grey, screen and printing colours 110 363 12 459
Changing style in the middle of a paragraph 110 563 12 682
Links and marks on placed text 110 681 12 816
Lists made from an indent and a tab stop 110 883 13 119
One pattern, repeated across whatever it fills 111 351 13 389
One drawing, placed many times 111 086 13 429
Breaking an overlong word 111 252 13 573
Fonts embedded in the file 111 279 13 691
A self-paginating table 223 298 27 499
The same numbers, said so they mean something 111 666 13 838
Internal and external links 112 095 14 117
Columns measured from their cells 111 935 14 111
A contents page with dot leaders 111 913 14 215
Tabs and tab stops 112 426 14 670
Cells merged down the rows 112 649 15 084
Overlong text in a narrow cell 112 900 15 328
A sheet that measures what it says 113 086 15 478
Colour the document defines 113 403 15 539
Custom page labels 113 543 15 792
A colour chart the drawing points into 113 810 15 970
One hatch, laid down in any colour you like 113 667 15 958
One colour, seen through four stencils 114 195 16 434
One rule instead of a hundred colours 114 356 16 799
A multi-page invoice table 114 981 17 195
Several plates at once, each at its own strength 114 968 17 361
A picture fitted into a rectangle 116 023 17 897
A sheet as it goes to the press 115 238 17 942
A handbook that says how it wants to be read 116 738 19 240
The device a file's colours are meant for 119 232 21 733
An archivable PDF/A-3 form 120 803 22 790
Factur-X: the electronic invoice 120 878 23 035
A file that explains its own entries 122 159 24 622
A register whose pages say where they came from 124 045 26 394
Pictures inside table cells 172 697 75 145
A gallery of picture formats 183 781 87 833
Images on a page 9 332 9 332

See the examples themselves

The speed figures on this site are measured with a licence key. Without one, every letter is drawn again as a shape: on a document made almost entirely of text, writing it was measured at about thirty-two times slower, and the file came out far heavier. A trial copy writes as many documents as you like. Every page comes out with a watermark: wording written across it. Under the watermark the pages are laid out exactly as a paid plan lays them out, with one difference: every letter is traced as a shape instead of being written as text. No text on the page can be selected, searched or copied; the file carries no font, and it is heavier than the same page written as text. Tracing a letter needs its glyph, so a page comes out only when every font its text is set in is either sent with your request or already held by the service. The editor on this site and the examples on the API page are set that way. Anything else needs a paid licence: the service creates no document, and its answer names what was asked for and says a licence key covers it. See what a trial copy creates.