Pixelate an image, or just the box you drag

Work back from the feature you want gone rather than from the size of the picture. A recognizable feature disappears once one block spans it, so measure the thing in pixels and use roughly a third of that: a 60-pixel face needs about 20, a line of small print needs a block as wide as two or three characters, and a whole-frame mosaic look starts around a fortieth of the frame width. The block count in the readout is the honest guide — under about 8 blocks across, you have a color study rather than a picture.

Because they averaged the sRGB bytes instead of the light those bytes stand for. An sRGB value carries roughly a 2.2 gamma, so the arithmetic mean of a black pixel and a white pixel is 128, which displays as a gray far darker than the halfway gray your eye expects — and every block containing an edge is exactly that case. The averaging here decodes to linear light, takes the mean and re-encodes, which is why the blocks sit at the same brightness as the untouched picture beside them instead of a step below it.

Yes — drag a box and only what is inside it is touched. Everything outside the rectangle comes through at full resolution, because the operation is a replacement of that rectangle's pixels on a copy of the original rather than a filter applied to the frame. If you need several separate areas covered in one pass, the censor page is the one for that — it holds a list of boxes rather than one.

Usually yes, and sometimes dramatically. A PNG of a heavily blocked picture compresses well because long runs of identical pixels are exactly what its filters were designed for, so a photograph that was 3 MB can land under 200 KB at a 24-pixel block. The exception is a light block over a noisy photograph, where you have removed almost nothing and added nothing, and the PNG lands larger than the JPEG you started with simply because PNG is lossless and JPEG was not.

Because the area almost never divides evenly by the block, and the remainder has to go somewhere. Splitting it across every block would shift the whole grid by a fraction of a pixel and blur every edge; dropping it would leave a strip of the original showing along two sides. So the final column and the final row take the remainder, which makes them up to one pixel wider or taller than the rest. The readout gives the mean rather than a round number for that reason.

A block that was more than half transparent stays clear, and the rest are painted fully opaque. The threshold matters when you pixelate a cut-out PNG: the ragged edge of the subject becomes a hard, blocky edge instead of a soft one, because a block is either kept or dropped and there is no half-block. If you want the transparency preserved smoothly, flatten the image onto a background first and pixelate that.

Not by itself, and this page does not claim it is. Pixelation replaces a block with its mean, which is a lossy summary rather than a deletion, and a summary of something short and high-contrast — a word, a number, a plate — can be tested against candidates until one matches. Use this page for the look and for softening detail nobody is hunting for. When the point is that something must not be readable again, take the redaction page, which flattens a solid fill into the exported pixels, or the plate page, which refuses the settings that leave a residue.