Complete Compression Reference

How to Compress a PDF File: Presets, Limits and Real Results

How to compress a PDF file to the exact size a portal will accept -- and how to reduce PDF file size when compression is the wrong fix. Five presets, target-size mode, real 2026 limits, all in your browser.

5 Compression Presets Target Output Size 100% In-Browser
By Karan Kumar Updated October 31, 2026 33 min read
Quick Summary

01Why Your PDF Got BIGGER After Compressing (And What to Do Instead)

The single most-asked question about PDF compression — and in most cases the file was never compressible in the first place.

PDF compression works by removing redundancy: downsampling oversized images, subsetting embedded fonts, stripping unused metadata, and re-encoding object streams. It shrinks files by making the bytes describe less. If a file is already minimal, there is no redundancy left to remove — and if a tool re-encodes it anyway, the output can be larger than the input. Nothing went wrong. The file simply had nothing to give.

The 5-second diagnostic — run it before you compress anything

Open the PDF, drag across a paragraph, and see what happens. If text highlights, you have a text-based PDF: there is almost no image data to downsample, and the savings will come only from fonts and metadata. If nothing highlights, or the whole page selects as one block, you have a scanned PDF made of pictures — and that compresses beautifully. Second check: divide the file size by the page count. A text PDF usually lands around 20–100 KB per page; a scan is often 300 KB to several MB per page.

What you foundWhat your file actually isWhat compression will doTypical result
Text highlights, 20–100 KB/pageBorn-digital. Text, vectors and fonts; no meaningful image payload.Metadata strip, object-stream repack -- no image work at all16–18% in our runs, identical at every preset
Nothing highlights, 300 KB–3 MB/pageA scan. Every page is a photograph of a document.Image downsampling and re-encoding25% at 150 DPI, 73% at 300 DPI (Medium) -- up to 97% at Extreme
MixedReal report: text plus charts, logos, photosOnly the image portion can shrink10–58% depending on the preset
Text highlights, but one huge image on every pageA chart or logo re-embedded at full resolution per pageImage downsampling -- this is the one text PDF that compresses wellNot measured here, but our photo deck lost 51–68%, and the image is the payload

Now the part that surprises people: rasterising a text page makes it bigger, not smaller. A page of crisp text is stored as a few kilobytes of text-drawing instructions. Re-render that same page as a picture at 150 DPI and you have created a 1275 × 1650 pixel bitmap — around 1.5 MB of JPEG data describing a page that previously cost 8 KB. This is why "compress my 40-page contract" often produces a file twice the size of the original, and why the honest answer for a text-heavy document is usually that compression is the wrong tool.

Reason the file grewWhat is happeningHow to avoid it
The file was already optimalEvery stream is already compressed, images are already at a sensible resolution. Re-encoding an already-good JPEG at a different quality level produces a bigger file.Run a pre-flight check before compressing — see below.
Text was rasterisedVector text and line art were replaced with a bitmap, which is far more expensive than the instructions that drew it.Use the object-level pass, which leaves vectors and fonts untouched.
Re-compressing a compressed PDFSecond pass over a file that already gave up its savings. The saving was taken on the first pass; the second can only cost quality.Compress once, at the right level. Do not re-compress a compressed file.
You asked for a forced transformGrayscale, PDF/A and Normalize deliberately re-encode every page, so they can grow a file that was already tight.Use those modes only when the output format matters more than the byte count.
How AFFLIGO avoids handing you a bigger file

The Compress PDF tool runs three checks before it hands you anything, and all three exist to prevent exactly this outcome.

1. A pre-flight scan before any engine runs. Before touching the file, the tool walks its raw bytes and measures two numbers: how much of the file sits in uncompressed streams, and how much sits in image streams. If less than 3% of the file is uncompressed and less than 50% is image data, the file is already optimised — the deep compress is skipped and your original is returned with the note "Already looks optimized (pre-flight check) — kept original, deep compress skipped." This happens in under a second and saves minutes of wasted work.

2. A fast object-level pass that never rasterises. The first engine is a quick pdf-lib + canvas pass that strips metadata and downsamples oversized images while leaving text and vector graphics exactly as they are. It finishes in seconds and, because it never turns a vector into a bitmap, it essentially cannot make a text PDF bigger. It only hands off to the deeper Ghostscript engine when it genuinely found something to fix.

3. A guard on negligible wins. The deep Ghostscript pass re-encodes every JPEG in the document, which costs a little quality. If the result saves less than 5%, the tool discards it and keeps your original, telling you the exact figure: "Only 3% smaller — kept original to protect quality." A 3% saving is not worth a generation of image loss.

Where the guard does not apply — be honest about it

Those three protections cover standard compression. Three modes deliberately bypass them, because you are explicitly asking for a different output rather than a smaller file: Grayscale, PDF/A, and Normalize all re-encode every page, and target output size keeps pushing until it hits the number you named. In any of those modes a text-heavy PDF can genuinely come out larger — that is the price of the transform, not a bug. If your priority is bytes rather than format, compress in standard mode first and apply the transform afterwards.

If compression is the wrong tool, use this insteadWhat it doesRealistic saving
Delete the pages you do not needA 40-page report cut to the 8 pages someone actually asked for, using Organize PDF to remove the restUp to 80% — and zero quality loss, because nothing is re-encoded
Split it into partsBreak the document into smaller files that each fit an upload or email limit, using Split PDFWhatever the limit was — this is the standard answer to a hard size cap
Pull the pages out as images and re-assemble them smallerExport each page with PDF to JPG at a lower resolution, then rebuild with Image to PDFUp to 70–90% on image-heavy documents — at the cost of a selectable text layer
Rebuild from the source fileExport fresh from Word, Google Docs or InDesign instead of compressing the exported PDFUsually the largest saving available, because you skip a lossy step entirely

02Do You Even Need to Compress?

Most places people share PDFs these days will happily take the file as it is.

WhatsApp and Telegram accept documents up to 2GB. Gmail's own limit is 25MB, and a file past that limit doesn't bounce -- it simply arrives as a Google Drive link. If you're sending a brochure to a colleague or dropping a form in a family group, compressing first costs you image quality for nothing.

Compression earns its keep in exactly one situation: the other side enforces a hard upload cap. Job portals, university applications, government e-filing, court systems and older corporate email gateways reject your file outright if it's a byte too big. That is where the rest of this guide helps -- and the next section matters more than you'd think, because the cap you're fighting is probably not the number you heard from a friend.

Rule of thumb

If the file already sends, leave it alone. Keep a compressed copy only when a specific portal demanded one -- and delete it when the application is done.

03The Real Size Limits (2026, With Sources)

Actual caps from actual services, checked in October 2026 -- because "keep it under 10MB for email" is advice from 2010.

Where you're uploadingThe real limitWatch out for
Gmail25MB total attachments (Google)Base64 encoding inflates attachments by about 33% in transit, so a 24MB PDF can still bounce. Past the limit Gmail sends a Drive link instead -- the recipient never sees a rejection, just a link.
Outlook.com25MB per message (Microsoft)Microsoft's general Outlook guidance still quotes 20MB for internet email accounts. When in doubt, 15MB clears both numbers with room to spare.
Yahoo Mail25MB across all attachments (Yahoo)The cap is the sum of every attachment in the message, not each file.
WhatsApp2GB per document (WhatsApp)You almost never need to compress for WhatsApp.
Telegram2GB free, 4GB Premium (Telegram)Same story -- leave the file alone.
LinkedInUnder 2MB recommended for resumes (LinkedIn)Worded as a recommendation, not a hard cap. PDF or Word both accepted.
Common App2,000KB per uploaded score report (Common App)Their own support article tells you to compress the PDF if it is bigger -- the limit applies per file, so a 2.4MB score report has to come down.
UCAS5MB per document, 5 at a time, 30 per application (UCAS)The batch limits trip people up as often as the size does.
India -- Income Tax e-Proceedings5MB per attachment, 10 attachments, 50MB total (Income Tax Department)This is the notice-response workflow. Photo and signature uploads on the same portal are separate, image-only fields with much smaller KB-range limits.
USCIS (US)12MB per file (USCIS)PDF, JPG or JPEG; some forms also take TIFF. Their guidance explicitly says do not encrypt or password-protect uploads -- and forms are tuned for PDF 1.7.
HMRC CDS (UK)10MB per file, 10 files at a time (GOV.UK)Customs declaration support documents specifically. Send more than 10 and you must split them into separate batches of 10 or fewer.
US federal courtsSet per court -- no national number (PACER)CM/ECF says each court sets its own PDF limit and points filers to the Court CM/ECF Lookup. If a filing bounces on size, break it into separate documents rather than guessing a number.
The Indian form trap

Photo and signature slots on UPSC, SSC and banking forms are image fields, usually capped at 10-200KB -- they want a smaller JPG, not a compressed PDF. If your photo upload keeps failing, compressing your PDF will never fix it. Shrink the image itself.

Portals change limits without notice. Treat the cap written in your own application form as the final word, and aim for about 90% of it -- landing at 99.9% of a cap is how files bounce on encoding overhead.

04How PDF Compression Actually Works

A PDF is not a picture of a page. It's a list of instructions -- which is exactly why some files shrink beautifully and others refuse to budge.

The file says "draw this glyph from this embedded font, place this image at this size." It doesn't store a photograph of the page. Compression attacks the parts that carry redundancy, in strict order of how many bytes they waste:

01

Measure before you touch anything

Before any engine starts, the tool walks the raw bytes and counts two things: how much sits in uncompressed streams, and how much sits in image streams. Those two percentages decide the strategy -- and if a file is already tight, they let the tool skip the whole job instead of grinding for nothing.

02

Downsample oversized images

This is where nearly all the bytes are. Images that are stored at a higher resolution than the preset asks for get resampled down: a 300 DPI scan at Medium becomes 150 DPI, which quarters the pixel count with no visible change on a screen. Our 300 DPI colour scan lost 73% at exactly this step.

03

Re-encode what's left

Resampled images are written back out -- JPEG (DCT) for photographs, Flate/ZIP for line art and flat colour. Every JPEG re-encode costs a little detail, and that is the real price of the preset you picked. Text and vectors are never rewritten into pixels by either engine, which is why words stay crisp and selectable.

04

Strip metadata and repack the objects

Document info, XMP metadata, thumbnails, unused resources and duplicate objects are dropped, and the surviving objects are repacked into compressed object streams. Invisible, and it is the only reason a text-only PDF shrinks at all: our typed CV and vector manual both lost 16-18% without any image work, because the quick pass did exactly this and Ghostscript never ran.

That order explains every result in this guide, including the strange ones. A scan is almost pure image data, so the preset you choose changes everything (0% to 97% on our samples). A born-digital document is almost pure instructions, so it gives up 16-18% and then stops -- and it stops at the same number whether you pick Low or Extreme, because the deep engine is never invoked.

05The Five Presets, Honestly

What Low through Extreme actually do to your file -- no mystery 1-9 scale, no "visually identical" promises. The savings column is measured on the eight test PDFs from section 10, not copied from a marketing page.

PresetImage resolutionMeasured savingHonest quality note
Low300 DPI0-18%Only helps files whose images are bigger than 300 DPI. A 150 DPI scan came out at 0% -- nothing to downsample -- and a 300 DPI scan at 9%. Print and archival use.
Medium (default)150 DPI5-73%Huge on scans (73%), almost nothing on images already near 150 DPI (5%). Sharp on any screen -- the right first try.
High120 DPI45-89%The biggest single step for image-heavy files. Use it when Medium did not clear the cap.
Ultra72 DPI53-94%Screen-only documents. Fine print starts to soften at 100% zoom.
Extreme50 DPI58-97%Last resort for hard caps. A 1.13 MB scan became 38 KB -- about 6 KB per page. Check every page before you submit.
Compressing without losing quality: what is actually possible

Text, fonts and vector line art are never turned into a bitmap by either engine, so words stay crisp, selectable and searchable. That part is genuinely lossless. Images are where the bytes and the detail live: every preset above Low re-encodes JPEGs at a different resolution, and re-encoding an already-compressed JPEG costs a little quality each time. If a tool promises "80% smaller, zero quality loss", it is describing a mostly-blank scan, not a rule. A 1.13 MB colour scan really can lose 97% and still be readable; a typed contract never will.

Start at Medium. If the result is still too big, jump to High before you touch Ultra -- in our runs that single step did most of the work on image-heavy files (10% to 45% on the mixed report, 5% to 51% on the photo deck). Only reach for Extreme when a cap leaves you no choice, and check the output with the Compare slider first. When you know the exact number you need, skip presets entirely and use target-size mode.

06Compress by Target Size

Enter the number the portal gave you instead of guessing with presets -- and know exactly how the tool decides when to stop.

AFFLIGO's By Size mode takes a target in KB or MB (1 KB upwards) and works backwards from it. You are not choosing a quality level and hoping; you are stating the constraint, and the tool finds the best quality that fits inside it. That is the difference between "compress and check" and "compress to 2MB".

What happens when you press Compress

1

The quick pass runs first, at about 150 DPI

In size mode the object-level pass targets roughly 150 DPI. If its output already fits inside your number, you get it back immediately. Measured: a 1.13 MB, 300 DPI colour scan aimed at 500 KB came back at 316.7 KB in 1.6 seconds -- the quick pass alone, no Ghostscript run at all.

2

If it does not fit, Ghostscript binary-searches

The smaller file is handed to Ghostscript, which walks a ten-step quality ladder from 50 DPI up to 300 DPI, at most eight attempts, always keeping the highest-quality attempt that stays under your number. Measured: the same scan aimed at 60 KB converged at 56.8 KB (23 seconds); a 231 KB mixed report aimed at 100 KB came back at 97.7 KB (31 seconds).

3

If no setting fits, you are told

You get the smallest attempt plus the message "Target missed on N file(s) -- showing closest possible." That is information, not a failure: it means the pages themselves cost more than your budget, and the next section explains what to cut instead of what to tick.

Two expectations worth setting before you type a number

It aims at the ceiling, it does not touch it. Output lands at or below your target, sometimes well below: 500 KB in, 317 KB out, because the next rung up overshot. That is deliberate -- quality is only spent when the budget forces it. If you need a fixed quality point rather than a ceiling, use By Level instead.

Aggressive targets cost time, not just detail. A 60 KB target on six scan pages took 23 seconds of real work, and every rung is a full re-distill. On a 200-page document, set the target and walk away rather than expecting a one-second result.

Who actually gives you a number

Email Attachments

Gmail caps attachments at 25MB, and Base64 encoding overhead means a 24MB PDF can still bounce -- enter 10MB as the target and leave room for everything else in the message.

Government Portals

Many government upload forms have strict 2MB or 5MB limits. Enter the exact limit as your target to maximize quality while meeting requirements.

Job Applications

LinkedIn recommends under 2MB for resumes; Common App caps documents at 2,000KB. Enter the number printed in the application form, minus a small margin.

Cloud Storage

When storage space is limited, compress large archives to a specific size to fit within your quota. Enter the remaining space as your target.

Target sizeWho asks for itHow to hit it in AFFLIGO
100KBGovernment e-filing, visa, ID-document portalsBy Size mode → 100KB. Best when the original is under ~2MB and mostly text; split out extra pages first.
200KBJob applications, CV uploads, short reportsBy Size mode → 200KB. 2-4 page resumes usually land here with readable text.
500KBLinkedIn, banking and HR portalsBy Size mode → 500KB. Keeps more image detail than 100-200KB targets.
1MBJob/university portals, email portfoliosBy Size mode → 1MB. If text looks soft, split out unneeded pages and re-run.
2MBUniversity uploads, tender documentsBy Size mode → 2MB. Our 1.13 MB scan never needed this much room -- but a 40-page one will.
5MBUCAS and most government forms, shared drivesBy Size mode → 5MB. Large budget, so the tool usually stops at the quick pass in a second or two.
10MBSafe-for-every-inbox zoneBy Size mode → 10MB. Clears Gmail, Outlook and Yahoo (all 25MB) with room for the rest of the message.

07How to Reduce PDF File Size: Do the KB-per-Page Math First

Thirty seconds of arithmetic saves you an hour of compressing files that were never going to make it.

Before you run anything, divide the target by the page count. That's your budget per page -- and it tells you instantly whether the goal is reachable:

budget per page = target size ÷ page count

ScenarioBudget per pageVerdict
100KB cap, 10-page typed report10KBNot reachable by compression. Our dense typed CV measured 32KB per page, and the presets moved it 16% -- not 90%. Cut pages or re-export from the source.
500KB cap, 40-page text contract12.5KBOnly if the pages are light. Our vector manual ran 11KB per page; dense prose runs 2-3x that and will miss.
2MB cap, 20-page 300 DPI scan100KBComfortable. Medium took a 300 DPI scan to 53KB per page in our run; High takes it to about 22KB.
500KB cap, 40-page 300 DPI scan12.5KBTechnically yes, practically no. Extreme hit 6KB per page on our sample, but at 50 DPI the text stops being text. Split the document or rescan at 150 DPI.
100KB cap, 3-page photo bundle33KBGrayscale territory. Colour photos won't fit; grey plus High might scrape through.

One caveat worth stating plainly: those per-page figures come from the eight test files in section 10. A noisy real scanner capture -- 300 DPI, heavy grain, no clean white background -- compresses less than our clean test page. Treat the numbers as an order of magnitude, not a promise.

When the math says no, stop compressing and start cutting: delete the pages nobody asked for (a 40-page report trimmed to 8 is an instant 80% cut with zero quality loss), split the document if the portal takes parts, or rescan at 150 DPI if you control the scanner. Fewer pages is the one lever that beats compression itself.

08Step by Step: Choose the Right Strategy

This guide is about the decision, not the buttons. Four steps, in the order they actually matter.

01

Diagnose the file before you pick anything

Drag across a paragraph. If the text highlights, it is born-digital and no preset will move it much -- the answer is fewer pages or a fresh export. If nothing highlights, it is a scan and compression has real work to do. Then divide the file size by the page count: 20-100 KB per page is text, 300 KB to several MB per page is a scan. That one number predicts the outcome better than any preset name. Section 01 has the full diagnostic; section 07 turns it into a per-page budget.

02

Choose preset, target size, or fewer pages

  • No hard number to hit? Use a preset and start at Medium. On a 300 DPI scan that is already worth 73%.
  • A portal printed a limit? Use By Size and type that number -- do not guess a preset when you were handed the ceiling.
  • Budget per page under about 10 KB? Neither mode will reach it. Cut pages or split the file first.

If the file already sends, stop here. Compression only earns its place when something rejects the upload.

Compression mode chips for Level and Size
03

Tick only the switches that change bytes

Grayscale is the one advanced option that reliably saves space -- it took our 150 DPI colour scan from 25% to 42% saved, and it is the best single move on a colour scan heading for a strict cap. Linearize, Normalize and PDF/A-2b are output-format decisions, not size decisions: any of them can make a tight file larger, and all three turn off the quick pass, so a job that took a second takes a minute. Section 09 takes each one apart.

Linearize Normalize and Grayscale checkboxes
04

Verify the output before you submit it

Open the result at 100% zoom and check what actually matters: small print, a stamp, a signature, a logo's exact colour. Aggressive presets soften fine detail first, and a portal that accepts the file but cannot read the scan is no improvement. If it looks soft, go one preset lighter and cut pages instead -- on everything except a pure scan, that trade is the better one.

The click-by-click tour -- where the buttons sit, what the queue shows, how Download All works -- belongs to the Compress PDF tool page. This guide stays on the decisions.

▶Shrink It Live — Animated Compress Demo

Hit play and watch a simulated run — upload a 1.13 MB, 300 DPI colour scan, set a 60 KB target, leave grayscale off, compress with Ghostscript, and download. About 25 seconds.

SIMULATED SCREEN RECORDING • 0:25
Ready — hit play0:00 / 0:25

09Linearize, Normalize, Grayscale, PDF/A-2b

Four switches -- and an honest warning: only one of them makes your file smaller.

OptionWhat it actually doesWhen it's worth ticking
LinearizeRestructures the file so page 1 loads before the rest arrives -- progressive display over the web. It adds a small hint table, so the output can end up a hair larger.A big PDF embedded on a web page where first paint matters. Never tick it hunting for megabytes.
NormalizeRewrites internal structures for compatibility with older or picky viewers. Size stays roughly the same; a very tight file can grow.The PDF will be opened on mixed devices, old readers, or strict government viewers.
GrayscaleConverts colour to grey before re-encoding -- and disables the quick pass, so the file goes through a full Ghostscript re-distill. Measured on our 150 DPI colour scan at Medium: 25% saved in colour (0.7s) became 42% in grey (17s). On the 300 DPI scan at High: 89% became 93%, at 65s.Scans, text documents, anything where colour carries no meaning. Expect a much slower run -- seconds to a minute instead of under a second. Stamps and signatures usually survive grey; check them in Compare to be sure.
PDF/A-2b (Archival)Converts to the PDF/A-2b archival standard: fonts and colour profiles embedded, interactive features stripped, conformance validated. The output usually gets bigger -- that's the point of archival.Courts, archives, regulators, long-term records. It's a compliance switch, not a size switch.
The honest combination

If your goal is bytes: pick a preset, add Grayscale if the content allows it, done. Linearize, Normalize and PDF/A are output-quality decisions -- enabling all four while hunting for megabytes works against you. And remember the guard from section 1: when a forced transform would have grown the file, the tool keeps your original instead.

10How Much Can You Compress? (Measured)

Eight files, five presets, forty runs through the live tool. Every number below is a measured result -- including the rows that barely moved, which are the most useful ones.

Test fileOriginalLowMediumHighUltraExtreme
Typed CV
2 pages, text + rules
63 KB16%53 KB16%53 KB16%53 KB16%53 KB16%53 KB
Vector manual
10 pages, line art
113 KB18%92 KB18%92 KB18%92 KB18%92 KB18%92 KB
Mixed report
12 pages, text + charts
231 KB10%207 KB10%207 KB45%127 KB53%110 KB58%98 KB
Photo deck
8 pages, full-bleed photos
174 KB5%165 KB5%165 KB51%85 KB61%68 KB68%56 KB
Colour scan, 150 DPI
6 pages
381 KB0%381 KB25%287 KB52%182 KB67%124 KB83%64 KB
Grey scan, 150 DPI
6 pages
363 KB0%363 KB23%278 KB51%178 KB66%122 KB83%63 KB
Colour scan, 300 DPI
6 pages
1.13 MB9%1.0 MB73%317 KB89%130 KB94%72 KB97%38 KB
Grey scan, 300 DPI
6 pages
1.12 MB9%1.0 MB72%316 KB89%130 KB94%73 KB97%38 KB

What the forty runs actually prove

01

The preset does nothing on text-only files

Look at the top two rows: identical bytes at Low, Medium, High, Ultra and Extreme. The quick pass handles those files and Ghostscript never runs, so the preset never gets a vote. Picking Extreme "to be safe" on a typed contract buys nothing and only adds a re-distill risk.

02

What matters is the source resolution, not the preset name

The two scans in the middle and the two at the bottom are the same six pages rendered at 150 DPI and 300 DPI. At Medium: 25% versus 73%. Same document, same tool, same click -- a 48-point difference created entirely by how the file was scanned.

03

Low is close to a no-op on modern scans

Low targets 300 DPI, so a file already at 150 DPI has nothing to downsample: both 150 DPI scans came back at 0%, five seconds of work for a file nine bytes smaller. On the 300 DPI scan it managed 9%. Low is for print archives and for images larger than 300 DPI, not for email.

04

The Medium-to-High step is the one that pays

Mixed report 10% to 45%. Photo deck 5% to 51%. 300 DPI scan 73% to 89%. If Medium did not clear the cap, High usually will -- and it is far kinder than skipping straight to Extreme. That is the single most useful line in this table.

05

Files already near the preset's resolution get a real result, just a small one

The 150 DPI scans still dropped 83% at Extreme, because re-encoding at 50 DPI is a genuine downsampling job. Nothing here is "already optimised" -- it is just that the safe presets had little left to remove.

How these numbers were produced

I ran these myself on the live tool and logged every byte -- none of them come from a marketing page. Eight generated test documents -- a dense two-page CV, a ten-page vector manual, a twelve-page mixed report, an eight-page photo deck, and one six-page scan rendered at both 150 and 300 DPI in colour and greyscale. Each run: fresh page load, one file, one preset, result read from the completed card so no previous run could leak into the next. Headless Chrome on Windows, same machine, same session. Timings are from that machine and will vary with your hardware -- a phone will be slower, a desktop faster.

What these numbers are not: a promise about your file. They are clean synthetic pages. A real scanner capture with heavy grain, a colour background or a photographed page compresses less at the same preset, because there is more entropy to preserve. Use the table to choose a direction, then check your own output in Compare.

11Scans, Photos, Text, Forms: What to Expect

Compression advice is content advice. What works on a scan does nothing for a contract -- and the measured numbers say exactly where the line sits.

Your PDF isWhat we measuredWhat to actually do
A scan at 300 DPI -- courts, offices, books73% at Medium, 89% at High, 97% at ExtremeThe best case there is. Medium first, then High. Grayscale adds a few points at High (89% to 93% in our run) but disables the quick pass, so it takes a minute instead of a second. Verify stamps and signatures in Compare.
A scan at 150 DPI or lower0% at Low, 25% at Medium, 52% at High, 83% at ExtremeSkip Low -- it targets 300 DPI, so there is nothing to downsample. Start at Medium, step to High.
A photo deck or brochure5% at Medium, 51% at High, 68% at ExtremeMedium can look like a failure here when the images are already near 150 DPI. High is the honest first try; check brand colours at 100% zoom.
Born-digital text -- contracts, papers, reports16-18% at every single presetStop compressing. Delete pages, split the document, or re-export from the source. That is where the megabytes are.
Text with charts and logos10% at Medium, 45% at High, 58% at ExtremeHigh clears most email and portal caps in our runs. This is the content type where the preset choice matters most.
A fillable formNot measured here -- test itAggressive modes can flatten fields or break their appearance. Fill after compressing, then type into two fields in the output to confirm they still work.
Digitally signedN/A -- compress before signingRewriting any byte invalidates the signature. Full explanation in section 13.

For scans specifically, remember the scanner is in your hands too. A 300 DPI colour scan is the most compressible -- and most oversized -- thing most people ever put in a PDF: 1.13 MB for six pages, about 193 KB per page. Scan at 150 DPI greyscale when the portal only needs legibility and you will rarely need a compressor at all.

12When the Portal Still Rejects Your File

You hit the cap and the upload still fails. Usually it isn't the size.

1

There's a minimum size too

Some photo, signature and visa slots refuse files under roughly 10KB. Over-compressed is a real failure mode -- if the file got tiny, compress less, not more.

2

Photo and signature slots are image fields

UPSC, SSC and banking forms want a smaller JPG in a KB range (often 10-200KB), not a PDF at all. The most common Indian-form rejection is uploading the wrong file type to the wrong field.

3

Page caps exist, separately from size caps

Some uploaders cap the page count and the byte size as two unrelated rules, so a 300KB, 40-page file can be refused for length alone. Trim pages before you touch quality. (The IRS Document Upload Tool is the one everybody searches for here -- its live help page lists no page limit, it requires the notice or letter number. Misfiled documents there fail on identification, not on size.)

4

Encryption gets silently rejected

Many portals refuse password-protected PDFs with a generic error. Remove the open password first -- Unlock PDF does it in your browser -- then upload.

5

The filename breaks it

Special characters, emoji and very long names trip up older uploaders. Rename to something plain: degree-certificate.pdf.

6

It's the session, not the file

Large uploads on slow connections time out mid-flight and the portal blames your document. Retry, or switch networks, before mangling the file further.

If all six check out

The problem is on the portal's side. Compressing down to 20KB hoping it slips through is how people end up with an unreadable document and the same rejection anyway. Contact the helpdesk with the exact error instead.

13What Compression Can Break

Four things a compressor can quietly damage -- and the ten-second check for each.

!

Digital signatures

A certificate signature is a hash over specific bytes of the file. Rewrite any of them -- which every compressor does -- and the signature shows as broken. Compress first, sign last. If a colleague already signed it, ask them to issue the final, smaller version; there is no safe way to re-compress a signed PDF.

!

Form fields

Heavy modes can flatten fields or break their appearance. Open the output, type into two fields, save, reopen. If the values vanish, re-run at a lighter preset.

!

The searchable text layer

Compressing a scan usually keeps its OCR text layer -- but verify: search for a word in the output. If your original was a scan with no text layer at all, run OCR before compressing, not after, so the searchable layer survives the downsampling.

!

Bookmarks, links and accessibility tags

On long documents, spot-check a few internal links and bookmarks after a heavy run. If your workflow needs accessibility (PDF/UA), check the tag tree survived too -- any mode that rasterises pages destroys it.

Password-protected PDFs compress fine, with one thing you should know: an open password is requested at upload, and permission locks (print/edit restrictions) are ignored. Either way the compressed file comes back as a plain, unencrypted PDF, because both engines rebuild the document from scratch and neither re-applies the encryption. That is helpful when a portal rejects encrypted uploads (see step 4 above) -- but if you need the protection back, re-apply it after compressing with Protect PDF, never before.

14How Small Can a PDF Actually Get? (Extreme and "Hyper" Compression)

Every site sells an "extreme" button. The honest version: there is a hard floor, and your pages set it, not the button.

Search for extreme compress PDF or hyper compress PDF and you will find tools promising 90% and up. That promise is real -- on one specific kind of file. Our 1.13 MB, six-page 300 DPI colour scan came out at 38 KB at Extreme, a 97% cut and about 6 KB per page. The same tool, same click, on a two-page typed CV: 16%. Nothing was broken. A scan is a stack of photographs and photographs compress; typed text is already a few kilobytes of instructions, and there is nothing left to remove.

File we measuredBest result we could getWhat that means for "extreme"
300 DPI colour scan, 6 pages, 1.13 MB97% at Extreme (38 KB), 89% at HighGenuinely extreme is available. Most of it arrives at High -- Extreme buys the last 8 points.
150 DPI colour scan, 6 pages83% at Extreme, 52% at HighStill large. The scan was already lean, so Extreme is doing real work rather than finding waste.
Photo deck, 8 pages68% at Extreme, 51% at HighBrand colours and gradients are the first casualty. Check them at 100% zoom.
Mixed report, 12 pages58% at Extreme, 45% at HighThe image half shrinks, the text half does not. 58% is the ceiling here.
Born-digital CV and vector manual16-18%, identical at every presetThere is no extreme setting for text. The deep engine never runs.

So the floor is roughly 6 KB per page on a scan at 50 DPI -- and at 50 DPI text stops being text. It stays legible on a screen, prints soft, and OCR may struggle with it. That is not a limitation of this tool; it is the arithmetic of a page image. If a portal demands 12.5 KB per page from a 40-page scan, extreme compression will technically reach it and the result will be barely readable, which is why section 07 sends you to page count instead of preset names.

If you genuinely need "hyper" compression

Do these in order and stop as soon as the file fits. 1. Delete or split out the pages nobody asked for -- free, lossless, and usually the biggest single cut. 2. Tick Grayscale if the document is not colour-critical. 3. Run By Size with the exact number the portal gave you, which searches for the best quality that fits instead of overshooting. 4. Only then drop to Extreme, and expect it to be slow: a 60 KB target on six pages took 23 seconds of real work, and every rung of the ladder is a full re-distill.

15Working Through Multiple PDFs

Queue as many PDFs as you like -- they compress one at a time, which is what keeps old laptops and phones stable. What matters at that scale is the order you work in.

Add multiple PDFs to the queue. They are processed sequentially to keep memory usage low, so browser-based Ghostscript stays fast and stable on every device, including older laptops and phones. Each file shows a linear progress bar and you can Cancel the current file. Finished files can be downloaded immediately or together via Download All (ZIP), and inspected with Compare.

Test on one file first

Run your worst file alone -- the biggest scan, the one with the photographs. Confirm the preset, then queue the rest. A batch that turns out to be one preset too aggressive is twenty files of wasted work.

Group by content, not by folder

Scans and text documents need different settings: a scan is worth 73% at Medium, a typed report 16-18% at any preset. Two batches with one setting each beats one batch you have to re-run.

Budget the time honestly

A 60 KB target on six scan pages took 23 seconds of real work, and every rung of the ladder is a full re-distill. Forty scans against a tight cap is minutes, not seconds -- start it and leave it. A preset run is far quicker than a target run.

16After Compressing -- What's Next

Workflow options after reducing your PDF file size.

Once your PDF is compressed, you may want to perform additional operations. AFFLIGO offers a full suite of PDF tools that work perfectly with compressed files.

Merge PDFs

Combine your compressed PDF with other documents. The Merge PDF tool lets you reorder and combine multiple files into one.

Protect PDF

Add password protection with AES-256 encryption. The Protect PDF tool secures your document before sharing.

Add Watermark

Stamp "CONFIDENTIAL" or your logo on every page. The Add Watermark tool supports text and image watermarks.

Convert to Word

Make the compressed PDF editable. The PDF to Word tool converts PDFs to editable Word documents.

Split PDF

Extract specific pages from the compressed document. The Split PDF tool separates pages into new files.

Rotate PDF

Fix page orientation if needed. The Rotate PDF tool corrects sideways or upside-down pages.

17Corrections to Earlier Versions

This guide has been rewritten around measured results. Here is exactly what changed, so you can trust the rest of it.

Every number on this page now comes from a run I logged on the live tool, or from the portal's own help page. That was not true of the earlier version, and the difference is worth spelling out:

Earlier version saidWhat was wrongNow
A preset table with "~10%", "35-45%" and "80-85%" savingsNot measured. It also implied text PDFs shrink 30-45% at Ultra.Forty logged runs across eight files (section 10). Text PDFs move 16-18% and then stop, whatever the preset.
"The IRS upload tool caps PDFs at 120 pages"I could not verify a page cap anywhere on the IRS Document Upload Tool help page.The row was removed from the limits table rather than kept with a footnote.
"The output keeps the same protection" for password-protected PDFsFalse. Both engines rebuild the document, so the password and permission flags are gone.Stated plainly in section 13, with the re-protect order to follow.
"Text and vectors are kept losslessly at every preset"Half true, and misleading as written: they are never rasterised, but metadata and object structure are rewritten.Section 05 separates what stays intact from what gets rebuilt.
An 8.4MB file shrinking to 1.9MB (77%) in the demoIllustrative, not a measurement -- and it pointed at the wrong tool URL.The demo now replays a logged run: 1.13MB to 56.8KB with a 60KB target, 95%.
"A by-size run targeting 150KB converged at 64,578 bytes"A single number presented as the rule. The tool lands at or under your target.Measured target runs in section 06: 500KB in, 317KB out; 60KB target, 56.8KB out.
"10MB clears Gmail (25MB) and Outlook 20MB"Outlook.com's own sending-limits page says 25MB, not 20MB.All three mail providers are listed at 25MB in section 03.
A "Common Mistakes" sectionIt repeated the diagnostic in section 01 and the breakage list in section 13.Removed. The page is shorter and says the same things once.

If you find a figure on this page you cannot reproduce with your own file, that is a bug in the page, not in your document. Tell me and it gets fixed the same way these did.

18Frequently Asked Questions

Answers to the most common questions about compressing PDF files.

Three reasons. (1) The file was already tight -- there was no redundancy left, and re-encoding an already-good file can cost bytes. (2) The document was rasterised, usually by an aggressive preset or a forced transform: a page of text is a few kilobytes of instructions, but re-rendering it as a 150 DPI picture costs around 1.5 MB of bitmap. (3) You asked for a transform that rebuilds every page, such as Grayscale, Normalize, PDF/A or a very small By Size target. Run the five-second diagnostic first: if the text highlights when you drag over it, it is a born-digital document and deleting pages will beat compression.
It depends on what is inside, and we measured it. In forty logged runs: a 300 DPI colour scan fell 73% at Medium and 97% at Extreme; a 150 DPI scan fell 25% at Medium and 83% at Extreme; a photo deck fell 51% at High; a born-digital CV and a vector manual fell 16-18% and then stopped, at every preset. Text PDFs have almost no image data to remove.
Medium (150 DPI) is the default and the right first try. If it does not clear the cap, step to High (120 DPI) before Ultra -- in our runs that single step did the most work on image-heavy files. Low only helps images stored above 300 DPI; on a 150 DPI scan it saved nothing at all. When you know the exact byte limit, skip presets and use By Size mode.
Use By Size mode, type the number your portal gave you in KB or MB, and press Compress. The tool runs a quick pass first and stops there if the result already fits; otherwise Ghostscript searches a ten-step quality ladder and returns the highest-quality attempt under your target. Measured: a 1.13 MB scan aimed at 500KB came back at 317KB in 1.6 seconds; a 231KB report aimed at 100KB came back at 97.7KB.
Because there is almost nothing in it to remove. A typed page is a few kilobytes of drawing instructions, and compression cannot shrink instructions much -- it is images that carry the megabytes. In our runs a two-page typed CV and a ten-page vector manual both stopped at 16-18%, identical at Low, Medium, High, Ultra and Extreme, because the quick pass handled them and the deep engine never ran.
Yes, and that is usually the point. The quick pass blanks the document info and XMP metadata, and the deep Ghostscript pass also drops thumbnails, unused resources and duplicate objects. That takes out author names, software stamps, original creation dates and often the GPS data a phone camera wrote into a page photo. It does not touch text that is part of the page -- a name or number printed on the page stays there. It is also not redaction: anything hidden under a black rectangle, or drawn in white text, is still in the file until you actually remove it.
Yes, but read this first: the compressed output comes back unencrypted. An open password is requested at upload and permission locks are ignored, because both engines rebuild the document and neither re-applies the encryption. That is useful when a portal rejects encrypted files -- if you need the protection back, re-apply it after compressing, not before.
Signatures, yes -- a digital signature is a hash over specific bytes, and every compressor rewrites those bytes. Compress first, sign last. Form fields and the OCR text layer usually survive downsampling, but verify: open the output, type into two fields, and search for a word. If your scan never had a text layer, run OCR before compressing.
Size is only one rule. Check for a minimum size (some ID and photo slots refuse files under about 10KB), page caps that are separate from byte caps, encryption (many portals reject protected PDFs with a generic error), unusual filenames, and whether the field wants a PDF at all -- photo and signature slots on Indian forms usually want a JPG in a 10-200KB range. The full checklist is in section 12.
Yes to both. Add as many PDFs as you like; they compress one at a time to keep memory low, and you can download each file or all of them as a ZIP. The same tool runs in a phone browser -- slower than a desktop, but there is nothing to install and nothing to upload.

Compress Your PDF Files Now

Take a 300 DPI scan down by up to 97%, or hit an exact size like 100KB or 1MB. Five presets, target-size mode, 100% in-browser -- free.

Open Compress PDF Tool

Key Takeaways

Karan Kumar, Founder of AFFLIGO

Karan Kumar

Founder & Web Developer at AFFLIGO

Karan built AFFLIGO's Compress PDF engine as a two-stage in-browser pipeline -- a quick object-level pass plus Ghostscript running 100% in the browser -- with five presets, a target-size binary search, grayscale, linearization and PDF/A-2b archival mode, and an inflate guard that keeps the original when nothing shrinks. Every number in this guide comes from forty logged runs of eight test files through all five presets, plus the target-size and grayscale runs quoted in sections 06, 09 and 10.

Last reviewed: October 31, 2026