What is the actual difference between a blur and a mosaic?

A blur mixes each pixel with its neighbours, so the detail is spread out rather than removed — a smaller version of the original is still in there, smeared. A mosaic replaces whole blocks with a single average, so within a block the detail is genuinely gone. A mosaic destroys more, which is why it is the better of the two, and neither of them destroys everything.

Is pixelation reversible?

Not in the sense of running an operation backwards to recover the original — the information inside a block is not there any more. It is reversible in the sense that matters, which is that the pattern of block averages can be compared against candidates until one matches. That works when the set of candidates is small and enumerable, which is the case for plates, dates, short numbers and any face drawn from a known group.

Which one looks best?

The blur, comfortably, and that is exactly the problem. The covering that looks most like a considered design decision is the one that protects least, and people choose with their eyes.

Can I use different coverings in the same picture?

Yes, and on a photographed document you usually should. Each region on this site carries its own covering and its own strength, so a solid bar over a reference number and a softer cover over a face can sit in the same frame and the same saved file.

Choosing

Blur, mosaic or solid

Three coverings, one rule for choosing between them, and a table for when the rule feels too strict.

The rule, before the table

Ask how many different originals could have produced what you are looking at. If the answer is a number somebody could count to, use the solid fill.

That is the whole decision. A face in a crowd of strangers has an enormous answer set — it could be any of billions of people and nobody has the list. A face in a photograph captioned with a school and a year has an answer set of about thirty, and a blur that leaves the hair, the height and the position intact narrows it further. The covering did not change; the context did.

The table

What is in the region	Use	Why
A stranger's face in a holiday photograph	Blur or mosaic, if you prefer how it looks	Nobody has the candidate list and nobody is going to build one. This is politeness rather than protection, and it is allowed to look good.
A child, anywhere	Solid fill	The cost of being wrong is not yours to accept, and the candidate set is almost always small — a class, a club, a family.
Faces in a work or school photograph	Solid fill	The group is named by the caption. A face-shaped smear of the right size in the right place is a strong hint when there are thirty candidates.
A number plate	Solid fill	Short string, known alphabet, standardised typeface, enumerable. The worst case in the subject.
A bank balance or an account number	Solid fill	Formatted, low-entropy, often check-summed. Half the answer set is gone before anyone looks at a pixel.
A signature	Solid fill	A signature is a shape, and a blurred shape is still a shape. It is also the one thing here that somebody might want to copy rather than read.
An identity number on a document	Solid fill	Same reasoning as an account number, with the added problem that the format is published and the covering is often applied to a large, sharp photograph.
A street name or house number	Solid fill	The answer set is a list of streets, which anybody can download.
A logo or a brand you would rather not advertise	Blur or mosaic	Nothing is at stake. Choose whichever looks less like a mistake.

Seven of nine rows say the same thing, and that is not a table that failed to be interesting. It is the finding. The interesting cases are the two at the ends.

Why the tool defaults to the boring answer

Most tools in this area default to a blur, because a blur is what the word in the search box said and because it photographs well in a product screenshot. The cost of that default is paid quietly and later, by somebody who assumed that covering a thing meant the thing was covered.

This one defaults to the solid fill and makes the other two a deliberate step down, with the reason printed beside the control rather than in a help page. It is a worse screenshot and a better tool. The full argument about what a blur actually leaves behind is in can blurred text be read.

Questions about the three coverings

What is the actual difference between a blur and a mosaic?: A blur mixes each pixel with its neighbours, so the detail is spread out rather than removed — a smaller version of the original is still in there, smeared. A mosaic replaces whole blocks with a single average, so within a block the detail is genuinely gone. A mosaic destroys more, which is why it is the better of the two, and neither of them destroys everything.
Is pixelation reversible?: Not in the sense of running an operation backwards to recover the original — the information inside a block is not there any more. It is reversible in the sense that matters, which is that the pattern of block averages can be compared against candidates until one matches. That works when the set of candidates is small and enumerable, which is the case for plates, dates, short numbers and any face drawn from a known group.
Which one looks best?: The blur, comfortably, and that is exactly the problem. The covering that looks most like a considered design decision is the one that protects least, and people choose with their eyes.
Can I use different coverings in the same picture?: Yes, and on a photographed document you usually should. Each region on this site carries its own covering and its own strength, so a solid bar over a reference number and a softer cover over a face can sit in the same frame and the same saved file.

The tool this was written for is on the front page.