Blur Like a Photographer: Background Blur, Privacy Blur, and the Tools That Do Each

A real portrait's background blur comes from a wide aperture: everything at the focus plane is sharp, and defocus grows with distance from that plane — a little for the subject's ears, more for the wall behind, maximum for the far corner. The visual cue that sells 'photo' versus 'filter' is that gradient. Uniform smear at one strength reads as damage to anyone who has looked at a real out-of-focus area, because real defocus is not a blanket — it is a slope.

So the professional setting is always lower than the first instinct. Blur to the point where the background stops competing with the subject, then stop — the moment it becomes a noticeable effect rather than an invisible one, it is too strong. A 15-25 px radius at full-frame scale with a distance gradient is a portrait; 60 px everywhere is a screensaver.

The clean method: remove the background first. Run the photo through the background remover to get the subject on transparent alpha, enlarge a blurred copy of the original to serve as the layer behind, composite (a design canvas, a slide, or the effects tool's own stack), and the blur is free and perfect because the subject is literally on a different layer. This is how apps without live depth sensing fake portrait mode.

The quick method: blur the whole image and mask. A uniform Gaussian over a photo whose background is already simple — sky, wall, water — reads as depth even though it is not graduated, because the subject's sharpness is doing the selling. This one works only when the scene has clean separation; blurring a busy room at the strength that would hide it muddies the subject's own edges.

Gaussian is the honest default: a true weighted average of neighbours, smooth and structureless, the blur behind almost every 'defocus' a browser produces. The blur tool exposes its radius; the entire art is choosing a smaller number than tempted.

Bokeh and radial presets add the lens's character — the slightly brighter rim where light bends around the aperture blades, the falloff from a focus centre. They look better than plain Gaussian on portrait scenes for the same reason real glass looks better than a box filter: light bounces, and the preset admits it.

Tilt-shift is the deliberate exception — a sharp horizontal band with rapid falloff above and below, borrowing the look of a view camera to make full-size scenes read as miniature models. It is the one blur type that asks to be noticed, and the only one where heavy strength is the point.

Pixelate is technically not blur at all: it replaces detail with blocks rather than smearing it. That difference matters for privacy (below) and for the look — pixelation is a style statement; Gaussian is an absence of one.

Blurring a face, plate or document before posting is a security decision, and weak blur fails: a 5-10 px Gaussian leaves enough structure that models can de-noise and reconstruct — a string of demonstrated attacks on 'privacy blur' shows face and plate recovery from mild smear. Pixelating hard, or covering with a solid shape, destroys the data instead of hiding it. When the image must read as unredacted, pixelate small and aggressively; otherwise opaque is the only fully defensible redaction.

The same rule flips for soft privacy — a café scene where passers-by should not be identifiable: 30-50 px Gaussian is enough, because the goal there is plausible deniability rather than cryptographic-level destruction. Know which goal you are buying.

One rarely-told upside: a blurred image compresses radically better than a sharp one, because JPEG and WebP spend their bits on high-frequency detail. A 4 MB hero background blurred lightly, saved at quality 70, can land at 150-300 KB and look better than a sharp version at the same bytes — the compression artefacts simply have no structure to attack. Blur + compress is the most underrated Core Web Vitals combination on image-heavy pages, provided the blur was an acceptable design decision to begin with.