What it means to pixelate image content
The effect replaces each square block of pixels with a single flat colour, so detail below the block size is destroyed and the overall shape survives. It is the standard way to obscure a face, a licence plate, an address bar or a name in a screenshot while leaving the rest of the picture readable.
The block size is the only real control, and it is the one that decides whether the tool did its job.
Averaging, not sampling
There are two ways to build this effect and they do not look the same.
The lazy way is one pass with nearest-neighbour scaling: for each block, pick whichever single pixel happens to fall at its corner and fill the block with that. It is fast and it looks wrong, because a single pixel is often not representative and the result comes out speckled and noisy.
The correct way is two passes. Scale the image down so one output pixel represents one block, with normal smoothing on, which averages every pixel in that block. Then scale that small image back up with smoothing off, so each averaged value becomes a hard square. This tool does the second thing, and it is why the blocks come out flat and even.
To censor image content, pixelation is not encryption
This is the part worth being blunt about, because plenty of people have leaked information they thought they had hidden.
Pixelated text is frequently recoverable. The reason is that the space of possible answers is small. An attacker who knows the font and the block grid can render candidate strings, pixelate them exactly the same way, and compare the result against your image. Postcodes, account numbers, short names and dates all fall to this quickly, and published tooling exists to do it.
The same weakness applies in a milder form to faces, where a known set of candidates plus a large enough block grid can still narrow things down.
So the rule is simple. If the hidden thing is merely private, pixelation at a generous block size is fine. If it is genuinely sensitive, do not obscure it at all: paint a solid rectangle over it, or crop it out of the frame entirely. A solid block has no information left to recover.
Choosing a block size to pixelate photo detail
Judge it against the output resolution, not against how the thumbnail looks. A 24-pixel block on a 4000-pixel-wide photo is a much finer grid than the same 24 pixels on a 600-pixel screenshot.
The honest test is to open the result at full size and see whether you can still tell who or what it is. If you can, go coarser.