Black out part of a video RECAST

Cover or pixelate, and why the default is cover

These are not two styles of the same thing. They are a real redaction and an approximate one.

Cover paints the area solid. Every pixel inside the box is overwritten with a single colour, so what was underneath is simply not in the exported file. There is nothing to recover because there is nothing left.

Pixelate divides the area into blocks and writes each block's average colour back over it. That destroys the detail inside a block, but it keeps a coarse picture of what was there, and for a small set of possible originals, a coarse picture can be enough. The attack is straightforward: take the font and the layout you can see, render every candidate string, pixelate each one the same way, and look for the one that matches. It has been used to recover redacted names, account numbers and licence plates from published documents and video more than once. Against a number plate, a postcode, a sort code, a short name, pixelation is a delay rather than a redaction.

Pixelate is here because it is sometimes the right answer: when you need the viewer to understand that a person is present, or where a face was, or that a screen was showing something, without showing what. If the goal is that the information is gone, use cover.

The box has to hold for the whole time it is on screen

A region applies over a stretch of the clip, and the stretch is yours to set: the whole thing, from a point to the end, or a window. The mistake that gets made is setting a box against one frame and not checking the rest. The camera pans, the subject leans, the document slides up the desk, and for eleven frames in the middle the thing you redacted is beside your box in full resolution. Nobody notices at normal speed. Everybody notices when the clip is stepped through.

So scrub the exported copy frame by frame across every range you redacted, at the edges especially, before you send it. A redaction that holds for ninety-eight per cent of its duration has not worked; the remaining two per cent is the whole of the information, and the export is now evidence that you tried to hide it.

What the export carries, and what it loses

The clip is decoded to frames, painted on, and re-encoded here in the tab. Because the output is built from pixels rather than edited in place, the original container is gone and the metadata that lived in it does not survive into the copy: no camera, no software, no location, no creation date. That is deliberate, and it is a second redaction you get for free.

The cost of the same mechanism is that the export is a re-encode, so it is a generation further from the original in quality, and the encoder your browser provides decides the format. If you need the original quality preserved and only the metadata removed, AVSCRUB strips it without touching the video. If the problem is what is being said rather than what is on screen, HUSH bleeps or garbles it and SNIP cuts the section out. For a still image the equivalents are VEIL for faces and plates and SCREENSHIELD for a screenshot, both of which cover rather than blur, for the reason above.