Convert MKV to JXL

JPEG XL, standardised as ISO/IEC 18181-1 in 2022, is unusual in offering both a strong lossy mode and a true lossless one in the same format, along with wide colour and high bit depth. That combination is why preservation collections and research image sets have started naming it in their policies: one format for the master and the access copy, with no argument about which variant is which.

It is a good reason to choose the format and it is the reason this page has to be careful. A policy that says "JPEG XL, lossless" is not satisfied by a file that happens to have a .jxl extension, and the difference is not visible from the outside.

Lossy JPEG XL, at a fixed quality setting, every time. The encoder here is called with a quality figure and without the lossless flag, and there is no control on this page to change either. The image is visually clean and it is not bit-exact with the frame that was decoded.

That is a legitimate output for an illustration, a reference, a working copy or a card on a page. It is not a preservation master and nothing about the filename says so. If your requirement is no loss, the honest routes are a PNG or a TIFF from this same page — both of which store exactly what was decoded — or a desktop JPEG XL encoder run in lossless mode on one of those files.

It is worth being precise about what lossless means for this material, because the word does more work in a policy than it does in a codec. A lossless still preserves the decoded frame exactly. It does not preserve the original scene: the video codec already discarded a great deal when the recording was made, and that discarding is permanent.

So a lossless frame from an MKV is a perfect copy of an imperfect thing, and its size reflects that — codec noise and blocking are close to random, and random data is what lossless compression cannot reduce. For an archive that is usually still the right answer, because the point of the policy is that no further decision is made downstream. It is worth knowing that the file will be large and that the largeness is not buying detail.

The image is written with no EXIF block, no colour profile, no source filename and no position in the recording. That is true of every still on this site — a decoded video frame has nothing to build those fields from — and it lands hardest on this reader, because a collection with a policy about compression almost certainly has one about cataloguing too.

The association therefore has to live outside the file, in the filename and in the record beside it. Naming each still with the source and the timestamp at the moment it is downloaded is a five-second habit that prevents an hour of reconciliation later, and it is the only place the information can go.

JPEG XL is read by fewer things than its merits deserve. Browser support is partial rather than universal, and outside the browser it is a handful of tools — GIMP, ImageMagick, a few viewers and converters — rather than everything that opens an image. Sending somebody a .jxl file is not a reasonable thing to do today.

For a stored archive that is a defensible risk, because the format is documented as an ISO standard and an archive expects to migrate its formats forward. For anything that has to be opened by a person on the other end of an email this week, it is the wrong choice, and JPG or PNG is the answer regardless of how much larger the file is.

The timestamp is the whole interface: seconds from the start, beginning at zero, accepting decimals and running up to a full day. On a long recording zero is a black frame or a slate, so the first useful action is to play the file, pause where you want it, and convert the elapsed time — one hour eight minutes is 4080.

Where the frame is placed changes the result more than any encoder setting would. A codec spends its bits on movement and describes the rest approximately, so a frame from a settled shot carries visibly more detail than one from the middle of a pan taken half a second earlier. For an archival still that is worth a second attempt at a slightly different second.

There is no resize control, so the still comes out at the recording’s own dimensions — 1920 by 1080 from a 1080p source, 3840 by 2160 from a 4K one. For most uses on this site that is more than anyone needs; for a stored copy it is exactly right, because the largest version is the one every later derivative is made from.

What is not available in either direction is enlargement. Nothing here can produce a still larger than the recording, and any tool that offers to is generating pixels rather than recovering them — which for a collection with an integrity policy is a worse outcome than a small image.

Only as much of the file as is needed to reach the timestamp is read, and both the decoding and the JPEG XL encoding happen inside the browser tab. Nothing carrying the recording is sent to a server, which for restricted collections, embargoed research material or anything under an access agreement is usually the condition that decides whether a tool can be used at all.

The free tier accepts an MKV up to 100 MB. Archival recordings are frequently far past that, and the practical route there is to export the frame locally with ffmpeg to a PNG and encode from that — which is also the route that gets you a genuinely lossless JXL, so for this reader the two limitations point at the same answer.