What we measured
Benchmarks first: where a claim here carries a number, it came off our own pipeline and the script that produced it is named. The explainers carry no numbers, and each one closes by saying what it did not test.
- Explainer8 min
The attachment limit is not the number you were told
Mail systems advertise one size and enforce a smaller one. Where the missing quarter of your budget goes, and what to do about a photo folder that will not send.
- Explainer9 min
What a converter keeps, strips, and quietly rewrites
Every image carries a second file inside it: location, camera, colour space, which way is up. Conversion treats each one differently, and only one of them is guesswork.
- Explainer8 min
PNG compression is lossless, so why does compressing a PNG lose something
Two different operations share one button. One rearranges bytes and changes nothing. The other throws away colours, and on a photograph it barely saves anything either.
- Practical7 min
Turning phone photos of paper into a PDF somebody can actually read
Most rejected document uploads fail for four reasons, and all four are decided before you convert anything: light, angle, resolution, order.
- Architecture10 min
We built the same converter twice, and the two disagree
Half of this site converts in your browser and half on our server. Measuring them against each other found the speed win we expected and three output differences we did not.
- 14.5s
- upload the browser skips
- black vs white
- same JPG conversion
- 177%
- worst size disagreement
- Measured9 min
Converting Word to PDF: what survives, and what leaks
Twelve generated documents through a headless LibreOffice. Font substitution turned out to be a solved problem. Tracked changes did not.
- 216.6pt
- identical fallback width
- YES
- deleted text in the PDF
- 1.2s
- warm conversion
- Explainer10 min
Why your browser still cannot open the photos your phone takes
Two separate reasons, and only one of them is about patents. What an iPhone actually writes into a .HEIC, and why our own server rejected every one of them.
- 25
- HEVC tiles per photo
- 16
- references allowed
- 2
- patent pools
- Benchmark11 min
I ran a real photo corpus through every format this site outputs
PNG, JPG, WebP, AVIF, GIF and PDF, measured on the same images with the same pipeline, including the PSNR that shows which ones actually kept the pixels.
- 46
- images measured
- 477
- megapixels
- +1,108%
- HEIC to PNG