PDF files grow large for two main reasons: heavy embedded images, and structural overhead left behind by the software that created them. This compressor attacks both, and is honest about which level does what.
Low rewrites the PDF’s internal structure losslessly with the pdf-lib library — pages, text and images stay bit-identical in appearance, and selectable text remains selectable. Savings here are real but usually modest.
Medium and High take a different route: each page is re-rendered as a high-quality image (150 DPI at Medium, 100 DPI with lower quality at High) and the document is rebuilt from those images. For scans and photo-heavy files this can cut the size dramatically — at the cost of turning text into pictures, so it is no longer selectable or searchable.