Choose text/vector-preserving optimization or separate raster compression.
Nerulio shrinks a PDF in two different ways. The default keeps every page object: images are resampled to the resolution the page actually draws them at and re-encoded as JPEG, embedded TrueType fonts keep only the glyphs in use, identical objects are merged and raw streams are compressed, so text, vectors and search are untouched. The optional "Flatten pages to images" mode instead renders every page into one JPEG: far smaller for scans, but the text can no longer be selected. When the result is not smaller, you get the original back.
Drop PDF files hereProcess many files at onceNo upload · processed on this device
Making a game?
Nerulio is a 2D game asset studio first: draw pixel art, cut sprite sheets into animations, pack texture atlases, check autotile tilesets and export them for Godot, Unity, Phaser and other engines. It runs in your browser; files stay on your device.
Two kinds of PDF compression and what each removes
In most PDFs the weight is in images and embedded fonts, not in the text. The structure-preserving mode works on those objects. For each image it finds the largest size at which any page draws it and caps the pixels at the level's resolution: Light 200 dpi, Recommended 120 dpi, Strong 90 dpi, together with a long-side cap of 2340, 1400 or 1050 px for files whose page size means nothing. A re-encoded image replaces the old one only if it is at least 5 % smaller.
Some objects are left alone on purpose: CMYK JPEGs, 16-bit images, stencil masks, images with a custom decode array, and fonts stored as CFF or Type 1 programs. Soft masks stay lossless. That is why a PDF made of vector drawings or already-compressed pictures can barely shrink.
The raster mode renders each page with PDF.js to the level's long-side pixel count and stores it as one JPEG. It removes everything a page is built from (text, vector paths, links) and keeps only its picture. That suits a scan that is already an image and does not suit a document someone needs to search or copy from.
Placement resolution
Image pixels ÷ the width the page draws it at, in inches. A 2400 px photo drawn 500 pt (6.94 in) wide is 346 dpi.
Font subsetting
Keeping only the glyph outlines the document uses; character codes and widths stay the same.
Rasterising
Turning a page into one image; text and vectors become pixels.
Example: the same pages at each level
Two typical images, and what each level writes back:
Photo 2400 × 2400 px drawn at 500 × 500 pt (6.94 in square)
Light 200 dpi × 6.94 in = 1389 px → 1389 × 1389 px, JPEG quality 0.82
Recommended 120 dpi × 6.94 in = 833 px → 833 × 833 px, JPEG quality 0.58
Strong 90 dpi × 6.94 in = 625 px → 625 × 625 px, JPEG quality 0.52
Scan 2339 × 3307 px filling an A4 page (11.69 in tall)
Recommended 120 dpi × 11.69 in = 1403 px, long-side cap 1400 px → 990 × 1400 px
Raster mode, Recommended, A4 page → one 990 × 1400 px JPEG per page, text gone
Pixel counts fall with the square of the resolution: the Recommended photo keeps (833 ÷ 2400)² ≈ 12 % of its pixels. A custom DPI (36–400, under Advanced) replaces the level's resolution and sets the long-side cap to DPI × 11.69, so 150 dpi means 1754 px.
Check the result
Read the result line: it gives the size before and after, how many images were optimised, fonts trimmed and objects merged, and whether text and search were kept.
Open the result and search for a word: in the default mode it is found exactly as before.
Zoom to 200 % on a picture that contains small print; if it is no longer readable, go back to Light or enter a higher DPI.
Troubleshooting
Symptom
Likely cause
How to check
Fix
The file barely shrinks
The weight is in objects this pass leaves alone (vector artwork, CMYK JPEGs, CFF or Type 1 fonts), or the file was already optimised
The result line reports few or no optimised images
Try Strong or a lower DPI; for scans only, the raster mode
"Already small — original kept"
The rewritten file was not smaller than the input
—
Nothing is wrong: you get the original instead of a larger copy
A scan made by an image-to-PDF converter shrinks less than expected
Such converters often declare one point per pixel, so the page looks huge and the scan seems to be 72 dpi; only the long-side cap applies
The page size shown by your reader is far larger than A4
Choose Strong or a lower DPI (which lowers the cap), and grayscale for black-and-white scans
Text can no longer be selected
The raster option was on
The result line says the text is not selectable
Run again with the raster option off; rasterising cannot be undone, so keep the original
Photos lost their colour
"Convert every image to grayscale" was on, or near-colourless pictures were stored in grey
The result line counts images converted to grey
Untick both grayscale options and run again
Other ways to do it
Option
When it is the better choice
Acrobat Pro's PDF Optimizer
When you need per-object settings and a report of which fonts, images or forms take the space; it is a paid desktop application.
Export again from the source application
When the PDF came from a word processor or layout program: a lower image resolution at export avoids compressing JPEGs twice.
When the limit is per file and the document is simply too long.
What this page does not cover
CMYK JPEGs, 16-bit images and CFF or Type 1 fonts are neither recompressed nor subset.
The raster mode removes text, links and vectors from every page; there is no mode that rasterises only the scanned pages.
Forms are flattened and certificate signatures are not preserved in the output.
Versions and sources
Checked in tests/pdf-browser.mjs on a 320-page document with 2400 px JPEG photos: the structure-preserving pass re-encoded the image objects, the file got smaller, text on the first pages and on page 320 stayed searchable, rotation and crop boxes survived, and the raster mode left no searchable text (Chromium 153, Firefox 155, WebKit 26.6). Savings on your own files will differ.
Choose text/vector-preserving optimization or separate raster compression.
How it works
Open a PDF and keep a backup
Choose preserve mode or raster mode, which removes searchable text
Compare the measured size and inspect text and images
Supported formats
PDF → PDF — Input decoding and output encoding depend on browser support.
What this tool does
Resample every embedded image to the resolution the page draws it at, trim embedded fonts to the glyphs in use and merge identical objects, keeping text and vectors.
Implementation and verified quality
advanced · placement-aware image resampling / reference-resolved recompression / TrueType glyph trimming / object dedupe / optional raster mode. Quality qualification is in progress. Pro/Flagship quality has not been established.
Before you export
Preserve mode retains native text, search and vectors; signatures are not retained. Font trimming drops the outlines of glyphs the document does not draw, and grayscale and raster mode are irreversible — raster mode removes searchable text. A forgotten password cannot be recovered, and unknown passwords are never guessed. The writer still parses full documents in a Worker.
Frequently asked questions
What determines the supported size?
Actual device memory, storage and codec support. Tested examples are evidence, not a guarantee for every file.
Are my files uploaded?
This tool processes the files you open on your device without uploading them. Images you explicitly attach to community posts are uploaded separately. Optional engines and models are downloaded from third parties; those requests expose connection information such as your IP address.
Will my work be saved automatically?
No. Images and editing state stay in this tab; media output can use local temporary storage. Changing language preserves your work; reloading, closing the tab, or following a link to another page can clear it.