Convert MKV to AVIF

Matroska is the container that ended up holding everything modern, because it is the one that never argued. HEVC at ten bits, AV1, VP9 profile 2, HDR10 and Dolby Vision metadata all travel in MKV files as a matter of routine, where MP4 handles several of them awkwardly and WebM is restricted by design.

So the person extracting a still from an MKV is much more likely than anyone else on this site to have a source with more than eight bits per channel in it, and much more likely to have chosen AVIF for exactly that reason. That expectation is reasonable, well informed, and wrong about this particular path — which is why it is the subject of the page rather than a line in the FAQ.

The frame is decoded and drawn onto a canvas, and the pixels are read back out of that canvas as eight bits per channel. Everything after that point — including the AVIF encoder — is working from eight-bit values. AVIF can store twelve bits and a wide gamut; there is simply nothing beyond eight left to hand it by the time it is asked.

Nor is a colour profile attached. The file carries the numbers that came off the canvas with nothing describing what colour space they belong to, so any viewer will treat them as sRGB. On a web page that is the right assumption and costs nothing. In a colour-managed pipeline it is the difference between a reference frame and a picture that looks approximately right.

If the still is a colour reference, a grading check or anything that has to represent the master, this is not the tool. A desktop decoder that writes straight to a high-bit-depth buffer — ffmpeg to a 16-bit PNG or an EXR, or a frame export from Resolve — keeps what the recording had and lets you decide how it is mapped rather than having a browser decide silently.

It is worth saying plainly because the alternative is a workflow built on a still that looks nearly right. Eight-bit values with no profile are perfectly serviceable for illustration and completely unsuitable as a reference, and the difference between those two uses is not visible in the file.

None of the above touches the reason most people reach for AVIF. It is a still image encoded by AV1, the codec designed to replace VP9 and compete with HEVC, and at the same visible quality it produces a substantially smaller file than JPG. That advantage is about compression efficiency and is entirely independent of bit depth.

So if the goal is the smallest good-looking still from a recording — for a page, a card, a preview, an app — this is the best target on the list, and the eight-bit ceiling costs you nothing you would have seen. The page above is for the reader who wanted something else and would not have found out until later.

AV1 searches far more possibilities per block than older codecs, and here it runs as WebAssembly on your own processor rather than on the hardware encoder your machine uses for video. A single frame from a 4K recording takes noticeably longer than the same frame as a JPG or a WebP — long enough to notice, not long enough to leave the desk.

It matters at volume rather than individually. One still is a pause; forty stills from a shelf of recordings is a real stretch of time, and that is the point at which taking the frames as PNG and batch-encoding them with a desktop AVIF encoder starts to be the faster overall route.

The timestamp is the only control on this conversion and it is the one that changes the result most. A frame from the middle of a pan, a fight or a crowd was described approximately by the video codec and no still encoder can recover what was never stored. A frame from a settled shot two seconds later can be dramatically better from the same file.

The box takes seconds from the start and begins at zero, which on a Matroska file is almost always a black frame or a logo. Play the recording, pause where you want it, convert the elapsed time to seconds, and type that in — an hour and four minutes is 3840.

Recent browsers read AVIF, which is what makes it reasonable to serve at all. The careful arrangement is still a `<picture>` element with an AVIF source and a JPG fallback, and it remains worth doing whenever a measurable share of your traffic is on older software you do not control.

Outside the browser the picture is thinner. Some desktop image viewers, several operating-system preview panes and a good number of upload forms still do not recognise the format, so an AVIF is a poor choice for anything that will be emailed, attached or handed to software somebody else chose. Keep a JPG for that and use the AVIF where it is being served.

Only as much of the file as is needed to reach the timestamp is read, and the decoding and the AV1 encoding both happen inside the browser tab. Nothing carrying the recording is sent anywhere, which for unreleased or graded material is generally the condition under which the work can happen at all.

The free tier accepts an MKV up to 100 MB. High-bit-depth 4K recordings are usually far past that, which is the awkward shape of this pair — the sources most likely to raise the bit-depth question are the least likely to fit. Where they do not, take a snapshot with a desktop player or export a PNG frame with ffmpeg, and convert that image here instead.