Compress a PDF by shrinking the images inside it

How the compression works

ToolGPT does not turn your pages into pictures. It reads the PDF, finds the JPEG images inside it, such as scanned pages and photos, and re-encodes them at a lower quality. Where an image has more pixels than a full page needs at 150 dpi, it is also scaled down. Everything else, including text, fonts and drawings, is copied over exactly, so real text stays sharp at any zoom and can still be selected and searched. In a scanned page, though, the words are part of the picture, so the stronger presets can soften them a little.

The presets trade size against picture quality. High quality changes the photos the least, Recommended is the balance, and Small file and Smallest compress harder and also drop the metadata. You can also ask for a specific image quality from 20 to 100.

When a PDF will not get smaller

Some documents give this tool nothing to work on. A PDF of pure text and vector graphics has no images to re-encode, and ToolGPT tells you so; at most it can remove the metadata. Only JPEG pictures are re-encoded. Lossless images such as screenshots and charts, JPEG 2000, black-and-white fax-style scans and CMYK print images are left as they are.

Encrypted, password-protected PDFs are refused. Every run is also measured afterwards: if the new file is not strictly smaller than the original, you get the numbers and no file.

Your document stays on your device

The whole job runs in your browser, so the PDF is not uploaded to a compression server. That matters for contracts, medical records and ID scans. The smaller copy is a new file with _small added to its name.