Hey folks!
The post may be a bit long. Please bear with me.
I am the developer of BentoPDF. It's a privacy first PDF toolkit that runs in your browser. I also have been working on making the best open source PDF compression engine, Hyper Compress, which is the what the post is about.
Why Hyper?
If you've ever needed to shrink a PDF you've almost certainly used Ghostscript. Even though it's great, but it rebuilds the whole document from scratch, so it can leave visual artefacts behind, and in our tests on 2,104 real PDFs it broke about 3% of files and made a quarter of them bigger.
Benchmark against open source tools
Hyper helps solve this issue. It is a high fidelity, monotonic compression engine. On those same 2,104 PDFs it beat Ghostscript, MuPDF and qpdf on size at every level, without making a single file bigger, and thus providing predictable outcome all while producing a conformant PDF.
Benchmark against commercial SDKs
For the new release I also ran a comparison against the commercial SDKs companies pay for. Hyper was tested against PDF Tools, Apryse and Nutrient on 1,000 real PDFs from public sources. The table shows size saved, with page quality after the @ (SSIMULACRA2, where 100 means identical to the original):
| Engine | Lossless | 200 dpi | 150 dpi | 72 dpi | |
|
|
|
|
| | Hyper | 18.0% @ 99.5 | 61.0% @ 89.8 | 70.2% @ 86.4 | 79.9% @ 67.9 | | PDF Tools | N/A | 55.1% @ 86.7 | 64.7% @ 82.9 | 75.6% @ 72.2 | | Apryse | N/A | 41.8% @ 88.4 | 55.6% @ 85.8 | 74.6% @ 73.4 | | Nutrient | N/A | 55.2% @ 87.8 | 61.5% @ 82.8 | 72.0% @ 70.8 |
Hyper made the smallest files at every level, and it's the only one with a lossless mode. Across its 4,000 runs it never failed and never made a file bigger. The others made between 5 and 120 files bigger, and Apryse and Nutrient failed on a few. At 72 dpi they keep a bit more quality, because they compress less.
But please Don't take my word for it. The list of PDFs with checksums, every per file result and the scripts are all in bench/sdk-comparison, so you can run it yourself.
It's available in CLI, Node SDK, C API, Docker, and can also run in your browser via WASM.
If you wanna quickly try it out: https://hyper.bentopdf.com/
It runs on your own browser via WASM, so nothing is uploaded.
I'd like to get more people testing it, especially with PDFs I haven't encountered myself. So if you work with PDFs regularly, I'd love for you to throw some difficult files at it and try and break it.
Any feedback and bug report is welcome and appreciated.
Thanks for reading and have a great day.
all 15 comments