Image Tools·4 min read

Pixelation vs Blur for Privacy: Block Size Matters

A grid of flat squares over a face is the internet's universal sign that something was deliberately hidden. The mosaic earned that status because it's simple and it mostly works, but pixelation vs blur for privacy is a genuine question with a genuine answer, and one slider decides most of it. Here's how the effect inside the Pixelate Image tool operates, how big your blocks actually need to be, and how the attacks that reverse mosaics really function.

Open the tool All insights

Where pixelation earns its keep: faces, plates, screens

The mosaic's natural habitat.

Bystanders and children in photos you want to post. License plates in a car you're selling online. Usernames and order numbers in app screenshots. These are the everyday pixelation jobs, plus the occasional purely aesthetic one, since a big-block mosaic doubles as a retro art style.

Mechanically the tool shrinks your image so each future block becomes a single pixel, then scales it back up with smoothing turned off, producing hard-edged squares. It processes the full frame; to mosaic just one area, crop it out with the Image Cropper, pixelate the crop, and paste it back.

Pixelation vs blur for privacy: what actually differs

Averaging into squares vs smearing smoothly.

Both effects throw away detail, but differently. A mosaic quantizes a region into a small number of flat colour averages. A blur spreads every pixel's value smoothly across its neighbors, leaving a continuous field that still encodes a lot of the original arrangement. At equal visual coverage, the mosaic generally retains less structure an attacker can exploit.

There's a social difference too. A mosaic unambiguously says something was censored here, which is often what you want. A blur can pass as depth of field, which suits backgrounds and product shots better than redaction.

Block size choices, translated into real numbers

Count blocks across the feature, not pixels.

The slider runs from 2 to 64 pixels, but the number that matters is how many blocks span the thing you're hiding. Block size choices should target fewer than 8 blocks across any identifying feature. A face 400 pixels wide needs blocks of 50 or more. A license plate 300 pixels wide at block size 16 still gets about 19 columns, enough for characters to remain guessable; at 48 it drops to roughly 6 columns and the plate is gone.

The live preview makes this concrete: drag the slider and watch the feature, not the image. When you can't tell a face from the back of a head, you're close.

Depixelation risks: how mosaics get reversed

The attacks need small blocks to work.

The classic attack is brute-force matching. If the attacker knows the font, layout, and character set, true for redacted account numbers and plates, they can pixelate every candidate string with the same grid and compare results against yours. Fine grids leave enough per-block variation to single out the right candidate. Machine learning reconstruction does something similar for faces, generating plausible identities consistent with the mosaic.

Both depixelation risks collapse as blocks grow, because each block is an average and averages destroy information. At 6 blocks across a face, thousands of different faces produce nearly identical mosaics, and that ambiguity is exactly the protection you're after.

A worked pixelation example, slider stop by slider stop

One portrait, three settings.

Input: photo.jpg at 3000x2000 with a face about 500 pixels wide. At block size 8 the face spans roughly 62 blocks and is trivially recognizable, just textured. At 16 it spans about 31 blocks, and acquaintances would still name the person. At 64 it spans under 8 blocks, reading as pure geometry. Output: photo-pixelated.jpg, same dimensions, with the original format preserved.

Pixelation mistakes that quietly undo the censoring

Redaction fails at the edges of the process.

The mosaic usually holds. What people do around it is what leaks.

Picking blocks small enough to look nice at full zoom. If the mosaic reads as a face at thumbnail size, it fails, because shrinking is the easiest attack there is.
Mosaicing text with a fine grid. Known character sets make matching attacks cheap; text deserves the largest blocks of anything you censor.
Leaving secondary identifiers untouched: reflections, tattoos, name tags, a distinctive jacket.
Sharing the pixelated copy while the original sits in a public album or an earlier upload.

Practical pixelation tips

Three habits worth keeping.

Size blocks relative to the feature, never to the image, since a small face in a large photo needs a coarser grid than the whole-image resolution suggests. Always do the thumbnail test by zooming the result out until it's small; if identity survives, raise the block size and re-run. For screenshot sets where the sensitive region sits in the same place, batch them together so every file gets identical treatment in pixelated-images.zip.

Pixelation and its companion redaction tools

The full privacy pass takes three tools.

The Photo Blur Tool covers the aesthetic side of hiding, backgrounds and soft focus, where a mosaic would look brutal. The Image Cropper enables the region-only workflow described above, and it's also the honest option when you can simply cut the sensitive area out of frame entirely.

Finish with the Image EXIF Viewer. This tool's canvas re-encode drops metadata from the output as a side effect, but verifying takes ten seconds, and GPS coordinates surviving in a censored photo would defeat the whole exercise.

Try it now

Open Pixelate Image

The tool is one click away. No sign up, no upload, no payment.

Open Pixelate Image

Write a review