Convert MKV to PNG

It is a common and slightly absurd situation: the talk was recorded and published, the slides were not, and the only copy of that diagram in existence is a rectangle on screen forty-three minutes into a ninety-minute file. Lecture capture systems and OBS both write Matroska, so the recording is an MKV whether or not anyone chose that.

The operation is simple and it is repetitive. One conversion produces one frame at one timestamp, so a deck of eighteen slides is eighteen runs with a different number typed each time. Each run only reads the file rather than re-encoding it, so this is a few minutes of work rather than an afternoon — but it is worth knowing the shape before starting, because past about thirty frames a desktop tool with an interval or scene-change option is genuinely the better instrument.

JPG was designed for photographs and it shows its assumptions on anything else. It stores colour at half the resolution of brightness and works in eight-by-eight blocks, which produces a faint ringing around every hard edge — and a slide is nothing but hard edges. Small text picks up a grey halo that makes it harder to read at exactly the size where it was already marginal.

PNG stores the pixels it was handed, without approximation. On a photograph of the audience the two formats are indistinguishable and JPG makes a much smaller file. On a slide, a spreadsheet, a terminal or a code sample, PNG is the difference between something you can read and something you can nearly read.

Lossless describes the step from the decoded frame to the file, and nothing before it. The video was compressed when it was recorded, the recorder chose a bitrate, and whatever it did to the small text on slide eleven happened long before this conversion existed. PNG preserves that faithfully. It does not undo it.

The practical consequence is that the recording’s resolution and bitrate are the ceiling, and no format choice raises it. A 720p capture of a slide with ten-point text produces an image where that text cannot be read, and the honest answer in that case is not a different image format — it is to find the deck, ask the speaker, or read the words off the video at full screen where motion helps the eye in a way a still cannot.

The timestamp is the whole interface, and where you place it changes the result more than anything else on the page. Do not take the frame at the moment the slide changes. A transition, a crossfade or an animated build is motion, and motion is exactly where a codec spends the fewest bits — the frame it stores there is a rough sketch that the next few frames correct.

Wait until the slide has been on screen and unmoving for two or three seconds. By then the encoder has had a keyframe or a run of quiet updates to work with and the text is as sharp as that recording will ever make it. The same rule explains why a still of a presenter gesturing is soft while a still of them standing still is not.

The box takes seconds from the start, and players show hours, minutes and seconds, so there is one multiplication per slide: hours times 3600, minutes times 60, added together. Forty-three minutes and twelve seconds is 2592. Decimals work if you need to nudge past a blink or a build step.

The efficient way through a deck is to scrub the recording once with a notepad open, writing down the time of every slide in one pass, and then to do all the conversions afterwards. Switching between watching and converting for each one is where the tedium comes from, and separating them turns eighteen slides into two short jobs instead of eighteen small ones.

A PNG of a slide is small for a lossless file, because flat backgrounds and solid text are the content lossless compression handles best — a 1080p slide often lands in the low hundreds of kilobytes. A PNG of the camera shot of the room is much larger, because photographic grain is close to random and random data does not compress.

Every image here is run through an optimiser after it is encoded rather than being written and left, which takes a real share off the size at the cost of about half a second. That half second is worth it: the file is the thing you are waiting for, and a directory of eighteen slides that is two thirds the size is two thirds cheaper to send to everyone else who missed the talk.

There is no resize control on this conversion. A 1080p capture gives a 1920 by 1080 PNG and a 4K one gives 3840 by 2160, and that is the whole story — the frame is decoded at native size and written at native size. For slides that is exactly right, because every pixel of a text frame is one you may need.

It also means the images can be larger than the document you are putting them in wants. Scaling them down afterwards in an image editor, or letting the destination do it, is fine and lossless in the direction that matters. What is not available is scaling up: nothing here or anywhere can invent detail the recording did not capture.

The decoding and the PNG encoding both happen in the browser tab. Nothing carrying the recording is sent anywhere, and for internal material — an all-hands, a training session, a session under an embargo — that is often the deciding property rather than a nicety.

The free tier accepts an MKV up to 100 MB. Lecture captures are routinely larger, which is the honest awkwardness of this pair: the files most likely to need slide extraction are the ones least likely to fit. For those, VLC will save a snapshot at the current position from its Video menu and mpv will do it with a keystroke, both working on the file where it sits — less precise about the timestamp, and free of any ceiling.