FyPDF — PDF & file conversion suite · Optimize

Pour the file smaller — without flattening the words underneath.

Most online compressors ship the cheap trick: flatten every page to a giant image, brag about the size drop, and quietly destroy the text underneath. The distillery does the slow version. Five stages, five presets, fonts subsetted instead of stripped, alpha and ICC profiles kept on the way through. Same document, lighter file.

Open the distillery Read the five pours

Text stays selectable
Compression never re-rasterises body type. Bullet points, paragraphs, and footnotes remain real text the next reader can copy or search.
Fonts subsetted, not killed
Embedded fonts get pruned to the glyphs your document actually uses — same metrics, same rendering, smaller payload.
Alpha & ICC profiles intact
Transparent images stay transparent. Embedded color profiles ride along so brand colors don't drift between original and compressed.
Reversible by intent
We never overwrite your source. The output is a separate PDF you can compare, accept, or throw out — and re-run with different settings.

charge · raw export.pdf

48.6 MB

01Object analysis

Read the document model

02Image recompression

Re-encode the rasters

03Font subsetting

Drop unused glyphs

04Stream optimization

Recompress content streams

05Metadata cleanup

Strip the cruft

yield · email pour.pdf

6.2 MB−87%

charge · 48.6 MB

yield · 5 stages · 6.2 MB

Yield−87%

The Distillery · 5 stages live

In-browserCloud-assist

The Five Pours

One engine, five honest trade-offs.

Every pour names what it changes and what it keeps — written down so the trade-off is on the menu, not in the fine print. Pick the one that matches the destination.

The Five Pours · Tasting MenuOptimize Track · 5 pours · Issue 04

No.

Pour

Target

Trade-off

Use

01Pour

Email pour

Send

TargetUnder 5 MB

The smallest pour that still survives an inbox attachment limit and renders crisp on a thumbnail preview.

Changes

Image DPI
96
Heavy JPEG
Aggressive subsetting

Keeps

Text
Vectors
Hyperlinks
Searchable metadata

02Pour

Web pour

Embed

Target−50% to −65%

Balanced for site embeds and viewer previews — small enough to load on slow connections, sharp enough for retina displays.

Changes

Image DPI
144
Smart JPEG/WebP
Mid-tier subsetting

Keeps

Vector type
Layered structure
Tag tree

03Pour

Print pour

Press

Target−25% to −40%

Modest reduction for press-bound work — lossless on vectors, careful on rasters, ICC profiles untouched.

Changes

Image DPI
300
Lossless re-encode
Stream packing

Keeps

ICC profiles
Bleeds
CropBox
Alpha
Vector strokes

04Pour

Archive pour

Store

Target−5% to −15%

Lossless re-pack only — strip the metadata cruft, recompress the streams, leave every glyph and every pixel untouched.

Changes

Stream re-pack
Junk metadata strip
Index rebuild

Keeps

Pixels (1:1)
All fonts
Annotations
Document tags

05Pour

Custom pour

Tune

TargetYour call

Open the panel, dial each stage. Set image DPI, font subsetting strength, ICC handling, alpha treatment, and metadata scope per file.

Changes

Per-stage controls
Preview before save
Per-document profiles

Keeps

Whatever you tell it to
Side-by-side preview before commit

Send · best for

Newsletter PDFs · Outbound proposals · Vendor invoices · Mobile inbox previews

Target · Under 5 MB

Embed · best for

Docs sites · Sample PDFs in viewers · Knowledge-base attachments

Target · −50% to −65%

Press · best for

Pre-press hand-off · Tactile deliverables · Brand-critical exports

Target · −25% to −40%

Store · best for

Long-term archives · Audit-ready masters · Legal originals

Target · −5% to −15%

Tune · best for

Repeating workflows · Per-team profiles · Edge cases the presets don't fit

Target · Your call

The Fidelity Manifesto

What the distillery never burns off.

Four promises sealed onto every pour. The cheap path is to flatten a PDF to a giant image and brag about the size drop — these are the four things you give up when you do that, written down so you don't have to find out the hard way.

SealPromise · 01

Text stays selectable

FyPDF never re-rasterises body text to shave bytes. Paragraphs come back as paragraphs, headings as headings, and the next reader can still copy a quote or search for a phrase. The cheap path is to flatten everything to JPEG and call it compressed — these stamps are the line we don't cross.

SpecVector text preserved · ToUnicode CMaps retained · Search index intact in viewers that support it.

SealPromise · 02

Embedded fonts are subsetted, not killed

Where a font is embedded, its in-use glyph set is preserved and the rest of the font file is dropped. New rendering uses the same metrics, the same kerning pairs, and the same hinting — your document looks the same, weighs less.

SpecSubset map per page · OpenType / TrueType features kept · Substitution tables intact for ligatures and accents.

SealPromise · 03

Alpha and ICC profiles intact

Image transparency survives the recompression pass. Embedded ICC profiles ride along, so brand reds stay brand reds and skin tones don't drift between source and output. Print pours preserve OutputIntents so the press driver gets the colour space it expects.

SpecAlpha channel preserved on PNG and TIFF inputs · Embedded ICC retained · OutputIntent kept on print pours.

SealPromise · 04

Outputs are real PDFs

The compressed file opens cleanly in Adobe, Preview, Foxit, browser readers — same document, smaller payload. It re-edits, re-merges, re-extracts in the rest of the suite without quality loss. No flatten-on-save, no zip-of-pages, no proprietary container.

SpecPDF/A-compatible · Re-editable in subsequent operations · Indexes in any document search · No proprietary wrappers.

Stages of the Pour

Read the engine one cut at a time.

Five stages, in order. Each pour can stop at any of them — Archive stops after stage 04, Email runs the full set. Below, what each stage actually does, what it preserves, and what it changes.

StageRead the document model

Object analysis

Before anything is reduced, the engine walks the PDF object graph: pages, content streams, resource dictionaries, font tables, image XObjects, structure trees. The pour can't shrink what it doesn't first understand, and skipping this step is what produces the broken outputs cheap compressors leave behind.

Preserves

Outline tree, named destinations, internal links
Tagged structure (PDF/UA accessibility tree)
Annotations, redactions, form fields
Embedded color profiles (ICC) and OutputIntents

Transforms

Builds an internal map of every reusable resource
Detects duplicated images / fonts across pages
Identifies orphan objects safe to drop later

TechniquesCross-reference walkContent-stream parseResource-graph build

Stage 01 of 05

Pages

Outline

Resources

Image recompression

Image XObjects get re-encoded with the codec best suited to the content: photographs to JPEG/WebP at the target quality, line art to PNG with palette reduction, scans to JBIG2 where supported. DPI is stepped down to the destination's actual rendering target — 300 for print, 144 for web, 96 for email — never above what the page will display at.

Preserves

Alpha channel where original was transparent
Per-image color profile (ICC) intact
Vector overlays not touched by the raster pass
Original aspect ratio and orientation flags

Transforms

Photograph encoders → JPEG / WebP at target quality
Line-art encoders → indexed PNG palette
DPI re-ranged to the destination's actual render size
Duplicated images deduped to one resource

TechniquesJPEG / WebP / JBIG2Palette reductionResamplingDedup index

Stage 02 of 05

Photo300

→

JPEG144 DPI

−71%

Photo300

→

WebP96 DPI

−84%

Line artlossless

→

PNGindexed

−42%

ScanGRAY

→

JBIG2JBIG2

−68%

codec map · DPI floor

StageDrop unused glyphs

Font subsetting

Most PDFs carry the entire font file even when the document only uses 87 glyphs from it. The pour reads which glyphs each text run actually references and rebuilds the font subset to contain only those — same metrics, same kerning pairs, same rendering, an order of magnitude less weight in the final file.

Preserves

Glyph metrics and kerning
Hinting tables for display rendering
OpenType features used by the document
Substitution tables for accent / ligature pairs

Transforms

Font tables pruned to in-use glyph set
Duplicate font references collapsed across pages
Embedded subsets re-encoded at minimum payload

TechniquesTrueType subsetOpenType subsetGlyph-index mapToUnicode CMap

Stage 03 of 05

Garamond Pro · in-use map25/80

Pruned · 55−69%

glyph subset · 25 of 80

StageRecompress content streams

Stream optimization

Content streams (the page-drawing instructions) are typically Flate-compressed with default settings. The pour walks each stream and applies the most effective filter chain for the data inside — Flate at higher levels for text-heavy streams, ZSTD where the reader supports it, RunLength for repetitive vector instructions. The result is the same drawing instructions, dictated more concisely.

Preserves

Drawing instructions and text positioning
Page dimension and rotation flags
Inline image positions and z-order

Transforms

Flate streams re-deflated at level 9
Repetitive instruction blocks RLE-collapsed
Font and image references unified across pages

TechniquesFlate / ZSTDRunLengthASCII85 stripObject-stream packing

Stage 04 of 05

page-1220 KB→78 KB

page-2312 KB→110 KB

page-3188 KB→64 KB

fonts96 KB→32 KB

images410 KB→142 KB

flate level 9 · stream re-pack

StageStrip the cruft

Metadata cleanup

Most PDFs accumulate metadata that nothing reads — software trails, encoding hints, abandoned linearisation tables, cross-reference fragments from earlier saves. The final pass strips junk metadata while keeping anything functional (XMP author data, permissions, structural tags) and rebuilds the cross-reference table cleanly so the output is the smallest valid file the spec allows.

Preserves

XMP author / title / keywords
PDF permissions and security flags
Structural tag tree (PDF/UA)
Document-level OutputIntents

Transforms

Soft-deleted objects pruned
Cross-reference table rebuilt as one stream
Linearisation hint table dropped or rebuilt
Encoding-hint cruft stripped

Techniquesxref rebuildlinearisationXMP retainJunk-tag strip

Stage 05 of 05

Kept

· XMP author
· Title / keywords
· PDF/UA tags
· OutputIntent
· Permissions

Stripped✕

Stale linearisation
Old xref fragments
Encoding cruft
Soft-deleted objects
Vendor trail metadata

xref rebuilt · single object stream · linearised

kept · stripped · xref rebuilt

Before · After

Four pours, honest yields.

Four real PDFs that arrive on the still. Charge weight on the left, yield weight on the right, the pour and reduction stamped in the gutter — and an honest takeaway under each.

Chargeinput

48.6 MB

Sales deck (raw export)

Hi-res hero photography on every slide, embedded full-fat fonts, untouched JPEGs at 600 DPI.

.pdf

Compress PDFImage-heavy product deck−87%

Yieldoutput

6.2 MB

Email-pour deck

Same fonts subsetted, photographs re-encoded at 144 DPI, identical text and vector overlays.

.pdf

TakeawayLands in the inbox in one go — and reads on a phone preview without pinch-to-zoom.

Chargeinput

82.0 MB

OCR'd annual report

Two hundred pages scanned at 600 DPI, OCR layer per page, deep colour images on the financials.

.pdf

Compress PDFScanned annual report−86%

Yieldoutput

11.4 MB

Web-pour report

OCR text layer kept verbatim, scanned images re-encoded with JBIG2, ICC profiles preserved.

.pdf

TakeawayLoads in the browser viewer, stays searchable, reads through OCR exactly the same on the way out.

Chargeinput

26.3 MB

New-hire pack

Twelve sections, dozens of duplicate logo images, three different fonts embedded in full.

.pdf

Compress PDFMulti-MB onboarding kit−84%

Yieldoutput

4.1 MB

Print-pour pack

Logos deduped to one resource, fonts subsetted, vector lines untouched, print bleeds retained.

.pdf

TakeawaySame kit lands ready for the print run and ready for the welcome email — one file, two destinations.

Chargeinput

8.7 MB

Mechanical spec sheet

Almost all vector — line drawings, dimension callouts, small embedded font set, almost no rasters.

.pdf

Compress PDFVector-heavy spec sheet−18%

Yieldoutput

7.1 MB

Archive-pour spec

Lossless stream re-pack, junk metadata stripped, vectors and fonts pixel-identical to source.

.pdf

TakeawayWhen the source is already lean, the pour is honest about it — small reduction, lossless guarantee.

Who works the distillery

Five regulars at the still.

The personas who reach for the distillery weekly — and the specific pours they choose. Find the closest match to the file on your desk this morning.

Persona · 01The communicator

Decks that survive the inbox cap

The team finishes a 60-page client pitch by Thursday night. Outlook caps at 25 MB; the export is 70. Compression has to land it under the line by morning without re-rendering every chart.

Preferred pours

Email pour · 60 → 4.2 MB · text + vectors intact
Custom pour · Tune image DPI per slide for next batch

Persona · 02The webmaster

Page-weight matters every render

Sample PDFs sit on the docs site, served on every viewer load. Each MB pulls Lighthouse scores down on every analytics report. Compression has to hold visual fidelity at retina but cap weight per file.

Preferred pours

Web pour · Sample docs · −62% weight
Custom pour · WebP-out · 144 DPI floor for retina

Persona · 03The print prepper

Modest reduction, no fingerprint on the colour

Press wants a PDF/X file under 30 MB — but every brand colour and every soft proof has to ride along. ICC profiles, alpha edges, vector strokes can't shift one shade between source and press.

Preferred pours

Print pour · −32% · lossless vectors · ICC kept
Custom pour · Stream re-pack only · no raster pass

Persona · 04The archivist

Long-term storage, byte-identical inside

Audit masters live in cold storage for seven years. Reduction has to be lossless — strip junk metadata and rebuild the cross-reference, leave every glyph and every pixel exactly as the regulator received them.

Preferred pours

Archive pour · −12% · pixel-identical · xref rebuilt
Custom pour · Disable image pass · keep all metadata

Persona · 05The mobile-first designer

Loads on slow connections, reads on small screens

Users open product documents on rural connections through low-end Android browsers. Compression has to chase mobile-first first — small files, vector-text-clean, JPEG/WebP for photographs, no decorative bloat.

Preferred pours

Email pour · Mobile-first · WebP rasters
Web pour · Embed in viewer · 144 DPI cap

Common Questions

Before you pour anything, a few honest answers.

Question Index

Q01 · 01 / 07

How small can you make my PDF?

It depends on what's inside. Image-heavy decks routinely come back at a third or less of their input size with no visible loss. Vector-heavy PDFs (spec sheets, contracts, line drawings) reduce more modestly — sometimes only 10–20% — because the source is already mostly text and instructions, not big rasters. The honest answer is the engine reduces what it can without lying about what it can't.

Distillery Reference · 01

7 questions in the distillery FAQIssue 04 · Optimize

Pour the file

The still is hot. Tell it which pour to draw.

Drop the file, pick a pour, take the result. Same engine across all five presets — fonts subsetted, ICC profiles honored, text-layer kept selectable, no flatten on the way out.

Open the distillery Tour every track in detail

Pour dispatch · 5 presets

Issue 04

All pours · in-browser default · text intact

One Suite · Seven Tracks · Twenty-eight Tools and CountingStart with the surface →