Reduce PDF Size for E-Filing Without Uploading Your Files

There is no national number. PACER says it directly: “Each court sets a limit on the size of PDF files submitted for electronic filing”, and it points filers to a court-by-court CM/ECF lookup. The caps we could read at the source:

Three things to watch in that table. The first is the envelope. San Diego (per order), San Bernardino (per submission) and Illinois (per envelope) cap the whole batch as well as each document, so three PDFs that each clear the per-file limit can still bounce together. The second is the word “under”: N.D. Cal. counts 1000 KB as 1 MB and rejects a file at exactly 50.0 MB.

The third is stale or conflicting pages. The Second Circuit’s clerk page says 10 MB and was last modified in 2016, its local rule still mentions 10 MB, and its CM/ECF court-information page lists 50.0 MB. When a court’s own pages disagree, ask the clerk. Other courts may have moved on too. The District of Delaware’s 2016 notice announces a raise from 5 to 35 MB, and the District of New Hampshire’s 2015 notice a raise from 5 to 10 MB. Both are still online, and neither tells you today’s limit. Courts do not always date the current one. Use the court’s CM/ECF page, or your e-filing provider.

Our working rule is to aim for 70% of the stated cap, a habit from our guide to PDFs that are too big to upload. A file at 49.5 MB against a 50 MB wall leaves no room for how a portal counts.

Size is the rule that rejects you at upload. The format rules decide whether the clerk accepts the document afterward, and they constrain how you compress.

Two of those rules bite on compression. Illinois and Texas bar encryption, so remove any password in your PDF editor before compressing; we have not tested how Smol handles a password-protected file. And the scan rows deserve a second look, because scanner settings alone can change file size by about 16 times. N.D. Cal. weighed the same 317-page document two ways and published the results next to its scanner settings:

That is about 16 times between two settings the court allows. If your court wants black and white at 300 dpi, scan that way: bitonal was the small end of the court’s own test. If you are stuck with a color scan that must stay at 300 PPI, converting it to grayscale with Ghostscript first kept every page at 300 PPI in our test and took the file from 5,017,560 to 1,720,854 bytes, 65.7% smaller (the scanned-PDF guide has the command). That is done outside Smol.

Because it is a stack of photographs of pages. In our test, the same three-page agreement came to 87,788 bytes as a text PDF and 5,017,560 bytes as a 300 PPI color scan: 57 times larger, with zero characters of searchable text, as measured in our scanned-PDF guide.

That is why the fix differs by file. A scan has no fonts to subset and nothing to merge, so resolution and color are the levers. A word-processor export has few images, so a preset has little to work on, though a 21-page text-only contract still fell from 116,995 to 74,731 bytes under Smol, to 56,321 under Ghostscript, and grew to 247,782 under Preview’s Reduce File Size (method and all five routes here).

If you hold the original Word file, export a PDF from it instead of printing and scanning. The Ninth Circuit, the U.S. Supreme Court and eFileIL each say as much in their own rules, and in our test the scan was 57 times larger than the text PDF and had no searchable text.

Smol’s five PDF presets are long-edge pixel caps with a JPEG quality attached: 600, 900, 1,200, 1,800 and 2,400 px at roughly quality 25, 45, 60, 75 and 88. They are not DPI targets, so the PPI you get depends on the page. This is the measured result on a three-page scan (2,550 × 3,300 pixels, so the 2,400 px Original cap already downsamples it) that was OCR’d first, by a tool other than Smol:

Every row kept all 1,405 words, because an OCR layer is text and image compression does not touch it. The same scan without OCR tells the other half: re-OCR accuracy was 96.8% at 82 PPI and 1.42% at 55 PPI. Read the order into that: OCR first, compress second. The full test and the one built-in tool that destroys the layer are in the scanned-PDF guide.

Page size changes the arithmetic, because the presets cap pixels and not inches. Legal-size exhibits are common in filings, and on them even Large drops below the floor:

So on legal-size or larger pages, treat Large as the floor you test rather than the safe default, treat Medium as a last resort, and check the output at 100%. If the type is not readable, the file is under the cap and still wrong.

Loose photos are a separate case. If you shrink image files before placing them in a PDF, Smol resizes images to fit inside 2000 × 2000 px by default, so raise or switch off that limit when detail matters, and the photographers’ delivery page covers image files in more detail. The preset behavior is explained at length in what the PDF presets actually do.

1. 1

   Find the cap and set your target at 70% of it

   Use the court’s CM/ECF page or your e-filing provider, and note whether the cap is per document, per envelope, or both.

2. 2

   OCR any scan at full resolution first

   Smol does not OCR, so use Acrobat or OCRmyPDF. On our scan, the OCRmyPDF defaults cut the file by 45.1% at 300 PPI (to 2,754,671 bytes) while adding the text layer. Adding --optimize 3 --jpeg-quality 70 took the same scan from 5,017,560 to 1,372,066 bytes, still at 300 PPI, before Smol touched it. Use one of these two commands:

   brew install ocrmypdf
   ocrmypdf exhibit-12.pdf exhibit-12-ocr.pdf
   ocrmypdf --optimize 3 --jpeg-quality 70 exhibit-12.pdf exhibit-12-ocr.pdf

   The 5,032,225-byte OCR’d source in the preset table below is a different file from these two outputs, so do not expect 624,925 bytes after this step.

   One catch: OCRmyPDF’s documentation says it writes PDF/A by default and that --output-type pdf turns that off. So this step may be what makes your file PDF/A, which some courts want and some do not.

   Press ⌘F in Preview and search a word you can see on the page. If it finds nothing, the layer is not there.

3. 3

   Drag the files, or the whole folder, into Smol

   Choose the Large preset, unless your court sets a dpi rule for the uploaded file (see the settings table). A mixed pile of PDFs and photographs can go in together, and Smol writes each result next to the original with a _compressed suffix. The original is never overwritten by default.

4. 4

   Read the per-file results

   Any file that would have grown stays as it was (the direct compress path). Exhibits that did not move may already be optimized, and those are the ones to look at by hand.

5. 5

   Check size, text and structure

   Get Info on each output against your 70% target. Search a word. Page through at 100%. If your court wants bookmarks or PDF/A, confirm them now with a validator, because we have not tested either after compression and the OCR step may be what created the PDF/A.

6. 6

   Clean metadata as a separate pass, and sign last

   Compression does not strip metadata (next sections). Apply any digital signature after compressing; we have not tested what a compression pass does to an existing signature.

Speed in our batch test: 100 copies of one 1.6 MB four-page scan (152 MB, a duplicated test set rather than a mixed real folder) took 12.2 seconds at the Small preset on an M2 Pro. The full write-up is in batch compressing PDFs on a Mac.

Split it. PACER’s own advice is “to break the document into separate, smaller documents”, and W.D. Wash. spells out how: anything over 100 MB “must be divided into smaller documents/attachments” and labeled, with its example being Exhibit 1, Part 1 and Part 2. The Ninth Circuit likewise requires oversized attachments to be divided. Florida points oversized material to an electronic storage medium the clerk can accept. Smol does not split PDFs, so use Acrobat or any PDF editor that extracts page ranges.

Check one thing before you cut. Splitting our twelve-page layout into three files produced 91.5 MB of output from a 34.8 MB input, because each part carried its own copy of every shared image. On a scan, where each page has a unique image, splitting changed nothing: 5,017,560 bytes in, 5,016,808 out.

• Respect the envelope. Splitting an exhibit into three documents does not help if the submission cap is the one you are hitting.
• Upload in small groups. N.D. Cal. says ECF is less likely to freeze if you submit no more than 8 to 10 files at a time, whatever their size.
• Check the judge’s standing orders. They can add rules about how exhibits are grouped or excerpted beyond what the court’s CM/ECF page says.

If you already pay for Acrobat for redaction and OCR, keep using its optimizer and skip Smol. The case for a second tool is a large folder of exhibits, or the colleague who should never need a command line. Our Acrobat comparison and seven-tool PDF test go further, and the Smallpdf comparison covers the upload question.

Smol runs on your Mac. It has no account and no server for your files, and it uploads nothing. The shot below shows Activity Monitor with zero bytes sent. That is a fact about where the bytes go. It is not a statement about any law, bar rule or engagement letter, and we are not able to tell you what your jurisdiction requires.

Compression is not metadata removal

Our 162,058-byte test contract fell to 57,377 bytes at the Tiny preset, 64.6% smaller, and every Document Info key and XMP tag survived. Comment authors and embedded attachments survive too. Smol has a separate strip-metadata operation for document properties. On our test file, one command from a free command-line tool restored the cleared properties in our test (the PDF metadata test has the method), so do not treat the result as permanent.

Metadata removal is not redaction

N.D. Cal. warns that metadata “can reveal anything that was contained in the file at any time” and that tools which “black-out, cover over or remove sections of text” can be undone “by anyone to reveal the text underneath” (the court’s redaction page). Smol does not redact, and its metadata stripping is no substitute for redaction. Use a real redaction tool, then check the output by extracting its text. The wider picture for sensitive files is on our confidential documents page, and the two sibling posts cover legal and HR documents and what compression leaves in place.

Smol is a one-time purchase: $29 for one Mac, $49 for the Team license (three Macs, seats managed from an admin dashboard), or $20 a seat for 5 to 500 seats on Enterprise, also managed from one dashboard. Updates are free and there is no per-file quota. See pricing.

Consistency is the point for a team. Write down the settings (Large preset, originals kept) and have each person set them on their own Mac, so a paralegal, an associate and the records clerk start from the same choices. A Finder Quick Action adds a right-click for the person who will never open the app.

A watched folder can compress new files automatically, but it fits legal work only with care. Scans must be OCR’d before they reach it, because compressing first can wreck OCR (in our test, re-OCR accuracy was 1.42% at 55 PPI), and we have not tested a scanner-to-folder flow. Use the Large preset, which keeps 164 PPI on a letter page. Watched folders can make already-optimized files larger, a defect we disclose in that post, so do not point one at a folder of finished filings.

Seven cases, and the first three matter most.

Your court needs more than a small file. If it requires PDF/A, bookmarks, OCR, redaction or Bates numbering, Smol supplies none of them. Acrobat Pro does (US$239.88 a year prepaid, per Adobe’s pricing page). If you need those, Acrobat is the right tool.

The submitted file must be 200 or 300 dpi or higher. Every preset caps the long edge at 2,400 px, which is 218 PPI on a letter page. If your court’s resolution rule applies to the file you upload rather than to the scanner, a Smol preset takes you below 300, and only Original clears a 200 dpi rule such as NYSCEF’s (Large, at 164, does not). Use the OCRmyPDF defaults, which kept 300 PPI in our test, convert a color scan to grayscale outside Smol (which kept 300 PPI and cut our test scan 65.7%), or rescan in bitonal.

You are cleaning one document and confidentiality is the point. exiftool -all= followed by qpdf --linearize is free and rewrites the file; the restore attempt we ran against it found nothing to restore. On Smol’s strip-metadata output, one command from a free command-line tool restored the cleared properties in our test (method here).

Your PDFs are text-only. Our presets work on embedded images. Ghostscript reached 56,321 bytes on the text-only contract where Smol reached 74,731.

You need an audit trail. If you need a hash log or a documented procedure, a drop target will not give you one. Some PDFs, from generators such as ReportLab, also compress by almost nothing under Smol’s presets, so measure rather than assume.

You need to split, merge or sign. Smol does none of them. And it runs on macOS only, so in a mixed office it covers the Macs and nothing else.

You are filing one document, once. Preview is free. Run it, then check the output for the lost-text trap in the comparison table above. If the result searches and fits, keep your money.