Blur vs pixelate for safely redacting a photo
How box blur and pixelation hide sensitive detail, when each is safe enough, and the settings that keep faces and text from being recovered.
What a box blur actually does
A box blur replaces every pixel with the average of the pixels around it inside a small square. The width of that square is set by the strength slider: a radius of 20 pixels averages each point over a 41 pixel window. Because the original values are merged into a single average and then thrown away, the fine detail that made a face or a word legible is genuinely gone from the saved file. The larger the radius, the wider the average and the less of the original structure remains.
How pixelation differs
Pixelation snaps a region into large uniform blocks, so a whole square of the image collapses to one flat color. Visually it looks blockier than a blur, but the effect on hidden information is similar: within each block the detail is averaged away. The practical difference is that a gentle blur can leave soft, recognizable contours, while coarse pixelation removes almost all shape at once. For text especially, big blocks can be more reliably unreadable than a moderate blur.
Why a light blur is not enough
The mistake people make is using a light blur that still hints at the underlying shape. Researchers have shown that weak or repetitive obscuring can sometimes be partly reversed, and even by eye a soft blur over a name can leave enough letter shapes to guess the word. If you are hiding something that matters, push the strength high so the region becomes an even wash of color with no discernible edges. Check the preview closely and zoom in before you trust it.
Settings that keep detail from coming back
Draw the region generously so it fully covers the sensitive area with a margin, since a tight box can leave a readable sliver at the edge. Use a high blur strength, toward the top of the slider, for faces, plates, and text. Export to PNG so no compression artifacts sharpen the boundary. Because this tool runs locally and discards the averaged pixels, the downloaded file contains no hidden original layer to recover, unlike some editors that merely draw a shape on top.