Convert MOV to JXL

A card of footage is unbrowsable. Filenames from a camera are counters, thumbnails in a file manager are frame zero of every clip — which on a camera is a lens cap, a hand, or somebody saying "rolling" — and the only reliable way to find a take is to open it and scrub. Multiply that by a season of shoots and the archive is effectively write-only.

One well-chosen still per shot fixes it. A folder of index frames named after the clips they came from turns four hundred gigabytes into something you can look at in a minute, and it is the thing every production either builds or wishes it had. This page is for making those frames; the format question below is about what to store them as.

JPEG XL was standardised in 2021 as ISO/IEC 18181 by the same committee that produced JPEG, and it is meaningfully better than what it replaces: roughly twenty to thirty per cent smaller than a JPG of comparable appearance on photographic content, with far better behaviour in the flat gradients that video frames are full of.

Over a thousand index frames that difference is worth real storage, and index frames are exactly the case where lossy compression costs nothing — you are looking at them to identify a shot, not to grade one. Where it would be the wrong choice is a frame you intend to work on, and that is a different page.

JPEG XL has a lossless mode and it is one of the format’s better features. This conversion does not use it. Frame extraction has one control — the timestamp — so the JXL is encoded lossily at a fixed quality of 82, and there is no way to raise or lower it.

For an index that is the right setting and the absence of a control is a convenience. For anything else it is a limitation to know about: a frame that has to be exact should be a PNG, and a frame you want to tune should be exported as PNG and encoded separately with a tool that exposes the quality. Nothing here is hiding a lossless path that you have not found.

JPEG XL is a better format than the ones people use, and that has not translated into being able to open it. Browser support is partial and has been contested for years; a lot of desktop software, most content management systems and a great many upload forms simply do not recognise the extension.

For an index that is survivable, because the files are for you and you can choose a viewer that handles them. It becomes a problem the moment somebody asks you to send one — a producer, a client, a colleague — and the answer has to be a conversion rather than an attachment. Convert one file and open it in the application you actually intend to browse the library in before you convert four hundred.

Zero seconds is almost never right. The start of a take is a slate, a hand, an out-of-focus pan or the operator settling, and an index built from frame zero of every clip is exactly as useless as the file manager’s thumbnails.

Pick a moment a few seconds in where the shot is what the shot is: the subject in frame, the camera settled, the light as it was. It takes one conversion to test and it is the difference between an index you use and a folder you built once. If a take has a distinctive moment, use that one — the point is recognition, not representativeness.

The still comes out at the video’s own resolution, so a 4K clip gives a 3840-pixel JXL. An index frame does not need that: at 1280 pixels a still is instantly recognisable, opens faster in a browser and costs a fraction of the storage.

There is no resize control on this conversion, so the resizing is a step in whatever tool you use to manage the library. It is worth doing, because the difference between a full-resolution index and a sensible one is roughly a factor of nine in area — which over a thousand frames turns a few gigabytes into a few hundred megabytes and makes the folder pleasant to scroll.

No metadata is written at all. There is no EXIF, no timecode, no date, no colour profile and no note of which clip the frame came from or at what second. A folder of index frames with converter-generated names is not an index.

The convention is the whole product. Keep the clip’s own filename and add the timestamp — the same base name for the still and the footage is what lets somebody go from the picture back to the file in one step. Doing that as you convert is trivial; reconstructing it later from four hundred images is not possible at all.

Each conversion decodes one video and writes one image. A shoot of forty takes is forty runs, which is a slow half hour and is genuinely how a lot of small productions do it.

Past that it stops being sensible. An archive of thousands of clips wants a script or a desktop transcoder that can walk a folder, take a frame at a fixed offset from each file, and name the output after the input. Use this page to settle what that offset and that format should be, then automate it — the decisions are the hard part and they are the same either way.

Each frame is decoded from the MOV and encoded to JXL inside this browser tab, using the codecs the machine already has. Nothing carrying the footage is transmitted, which the network tab shows during a conversion.

For an index that is the practical point as well as the principled one. Building a browsable catalogue of a shoot means touching every file on the card, and doing that through a hosted converter would mean uploading the entire shoot to obtain a few hundred kilobytes of pictures. Locally, the size of the source costs decoding time and nothing else.