About the bench

There are four or five good tools for turning one named format into another named format. None of them helps the person who has not decided yet, and that person is usually reading an article about somebody else’s photographs while holding one of their own.

The useful thing to do for that person is not to write a better article. It is to run the comparison on their file and hand them the numbers, which is a categorically better answer than an average over a corpus they are not in.

Measured: every byte count on this site, read from a file the browser actually produced; every quality setting, found by search rather than chosen; every similarity score; every encode time, timed on the device it ran on. None of those numbers is a lookup.

Judged: the three thresholds behind the quality control, the margin at which a better-supported format outranks a slightly smaller one, and the test that decides a picture was drawn rather than photographed. Those are opinions with numbers attached. They are printed under the results so that anyone who disagrees can see exactly which one produced the recommendation and discount it.

That the similarity score is how a picture looks. It is a statistic that correlates with how a picture looks, better than the alternatives that will run in a browser tab, and worse than your own eyes at 800 per cent — which is why the panes are there.

That a format is better than another format in general. The bench has nothing to say about your next picture. It has a great deal to say about this one.

That converting is always worth doing. When the file you opened is already the best version of itself, the page says so and offers nothing to download.

Static files, a worker thread and no server. There is no account, no upload endpoint and no database, because none of those is needed to do this work and each one would be a place for a picture to end up. The advertising and the analytics are described in the privacy policy in the detail they deserve.