Blur the Background of a Photo

Look at any portrait taken by a working photographer and you will notice the same thing: the person is sharp and everything behind them has melted into soft shapes and colour. The effect is called shallow depth of field, and it comes from a large lens aperture that focuses on a thin slice of the scene. Everything in front of and behind that slice is out of focus, and the further away it is, the more it blurs. The eye is drawn irresistibly to the sharp subject because there is nothing else in the frame to compete with it.

Phone cameras have small lenses and tiny sensors, so they capture almost everything in focus at once. That is great for landscapes and terrible for portraits: the cluttered kitchen, the parked cars, the strangers in the café all stay as sharp as the person you were photographing. Manufacturers responded with “portrait mode”, which estimates depth and blurs the background in software — but only at capture time, only if you remembered to switch it on, and only on the photos where the phone’s depth estimate happened to work.

A background blur tool lets you add the effect afterwards, to any photo, from any camera. The otherwise perfect shot ruined by a busy background, the product photo taken on a messy desk, the old picture from a compact camera that never had a portrait mode: all of them can get the look. The key question is how the tool decides where the subject ends and the background begins — and, increasingly, where that decision is made.

Blurring is easy; knowing what to blur is hard. The tool needs a mask that says, for every pixel, how much it belongs to the subject and how much to the background. Twenty years ago that mask was drawn by hand with a pen tool. Today it comes from a segmentation model: a compact neural network trained on millions of photos to recognise people, pets and objects and to produce a soft mask with proper partial transparency at the edges — individual hair strands, the fuzzy edge of a jumper, the translucent rim of a glass.

The interesting part is where that network runs. Most online tools upload your photo to a server, run the model there, and send the mask back. The tool described here does the opposite: the first time you press Detect the subject, it downloads the segmentation model — about 45 MB — into your browser, where it runs on your own graphics hardware through WebGPU or WebGL. The editor tells you this is happening, and the download takes well under a minute on a normal connection. From then on the model is cached, and every later photo is detected in a second or two without any download at all.

The consequences follow directly. Your photo never leaves your device, because there is nowhere for it to go: the model, the mask and the blur all live in the browser tab. There is no queue and no counter, because nothing is costing anyone a server bill. You can load the editor, switch off the Wi-Fi, and keep detecting and blurring — the simplest possible proof that the processing is genuinely local. And because it is local, it is unlimited: an album of fifty portraits costs exactly the same as one.

The whole workflow is three decisions — detect, blur, adjust — and takes under a minute once the model is cached. Here it is in order, with the choices that matter at each step.

• Load the photo: drag it in, choose it with the file picker, or paste it from the clipboard with Ctrl+V.
• Open the Background panel and press Detect the subject. On first use the 45 MB cut-out model downloads into your browser; afterwards it is cached and detection takes a second or two.
• Check the mask. The subject is detected on your device; hair, ears and shoulders should have soft, natural edges.
• If an edge is wrong — a strand of hair classified as background, a piece of the wall stuck to a shoulder — open the AI brush, use Paint and Erase to correct the region, and come back.
• Choose Blur the background. The scene behind the subject softens with a smooth, lens-like falloff.
• Drag Blur amount. A gentle setting reads as a real lens; a strong one turns the background into abstract colour. Judge it at Fit to screen, not zoomed in.
• Hold Compare to see the original and confirm the subject itself is still perfectly sharp.
• Optionally finish with Light and Colour: a slight Shadows lift on the face, a touch of Vignette to draw the eye inward.
• Open Export, choose JPG, PNG or WebP, and download at the full resolution of your original. Nothing was uploaded at any point.

The most common mistake with background blur is too much of it. A real lens produces a gradual falloff: objects just behind the subject are slightly soft, objects far behind are very soft, and the transition is continuous. A blur that is cranked to the maximum flattens all of that into a uniform smear, and the eye registers the result as “edited” before it registers the subject. The Blur amount slider is there to be used in its lower half more often than its upper half.

A useful test is to ask what a photographer with a good portrait lens would have produced in the same room. In a small space with the background a metre or two behind the subject, the answer is a moderate blur where you can still tell there is a bookshelf, but cannot read the spines. In a park with trees thirty metres away, a stronger blur is plausible, and the leaves can become soft discs of light. Match the blur to the implied distance and the effect stops looking like an effect.

Consider the content of the background too. Plain walls need very little blur — they are already featureless, and a small amount simply hides texture and shadows. Busy backgrounds full of small high-contrast detail — a crowd, a shelf of products, a street — need more before they stop competing with the subject. And if the background is genuinely ugly rather than merely busy, remember that the same panel offers Solid colour and Transparent as alternatives: sometimes the cleanest portrait has no background at all.

Segmentation models are very good and not perfect. The places they struggle are predictable: fine hair against a similar-toned background, the thin frames of spectacles, a hand held away from the body, a pet with fur the same colour as the sofa, and anything the subject is holding that the model does not consider part of them. When the mask is wrong in one of these places, the blur either eats a bit of the subject — a soft ear, a smeared strand of hair — or leaves a sharp halo of background clinging to their outline.

The fix is the AI brush, working as a mask editor. Zoom into the problem area, choose a small Brush size, and use Paint to add missing parts of the subject or Erase to remove background that was misclassified. The Select the subject button gives you the detected mask as a starting point, and Invert flips it if you would rather paint the background than the person. Because this is mask correction rather than generation, it stays entirely on your device and costs nothing from your daily AI allowance.

For very fine hair, resist the urge to trace every strand. A soft mask edge that blurs partway through the frizz looks more natural than a hard edge that tries to be exact, because that is what a real lens does too — the outermost hairs of a real portrait are themselves slightly soft. Fix the obvious errors, check the result at Fit to screen rather than at 400 per cent, and move on.

Background blur is not only for faces. Product photos for a marketplace listing benefit enormously: a vintage lamp on a cluttered table becomes a lamp floating in soft warmth, and the buyer’s eye goes straight to the item. Sellers on eBay, Vinted, Etsy and Depop compete in a grid of thumbnails, and a listing with a clean, softly blurred background reads as more expensive than the one beside it with a radiator and a power strip in focus.

Pets are natural subjects — a dog on a walk with a park blurred behind it looks like a studio shot — and the segmentation model handles animals well, though fur edges may want a moment with the brush. Food photography works too: blur the rest of the table and the plate becomes the hero. For profile pictures, the recipe is simple and repeatable: crop to Square 1:1 with the eyes in the upper third, detect the subject, add a gentle blur, lift Shadows a touch so the face is open, and add a subtle Vignette. Five minutes, no photographer, and the same look across LinkedIn, Instagram and your email signature.

A blur that looks off is almost always a mask problem or an amount problem, and both are quick to fix. Hold Compare, look at the outline of the subject at Fit to screen, and run through the list below.

• The first detection is taking a long time: the 45 MB model is downloading into your browser. This happens once; afterwards it is cached and detection takes seconds.
• A sharp halo of background clings to the subject: the mask is slightly too generous. Open the AI brush, zoom in, and Erase the strip of background along the edge.
• Part of the subject is blurred — an ear, a hand, a strand of hair: the mask missed it. Paint it back in with the AI brush and reapply the blur.
• The background looks like a smear rather than a lens: Blur amount is too high. Bring it down until objects are soft but still legible as shapes.
• Two people, only one detected: paint the second person into the mask with the brush, or crop so the intended subject fills more of the frame and detect again.
• The subject itself looks soft: it may simply be soft in the original. Compare to check, then use Sharpness or Clarity in Detail on the result.

Background blur is one of the few AI-flavoured editing features that can be completely free without a catch, and the reason is architectural. The segmentation model runs in your browser, the blur is rendered by your graphics card, and the export is assembled on your device. No server is involved at any stage, so there is no cost to recover, no reason to add a watermark, no reason to downsize the export, and no reason to ask for an email address. The tool is free because it costs nothing to run — for one photo or a thousand.

This also means background blur does not touch your daily AI allowance. That allowance — twenty operations per visitor per day, shared with the image generator on the same site — is reserved for the generative operations that run on a server: Remove, Replace, Enhance, Restore, Colourise and Cartoon. Detecting the subject, blurring, replacing the background with a solid colour, making it transparent, and correcting the mask with the brush are all on-device and unlimited. You can blur an entire album without the counter moving.

Resolution is preserved because there is nothing to preserve it from: the blur is rendered at the native size of your photo, and the export is the same dimensions you loaded. Choose PNG if you have made the background transparent and need a cut-out, JPG for a finished photo going to a website or a message, or WebP for the best size-to-quality ratio on the web. EXIF metadata is stripped by default so that the GPS coordinates of your living room do not travel with your portrait, and one checkbox keeps it if you prefer.

“Is it really free and unlimited?” Yes. Detection and blur run entirely in your browser, so there is no limit, no account, no watermark and no reduced-resolution export. “Is my photo uploaded?” No — the subject is detected on your device by a cached model, and nothing leaves the browser. “Does it work offline?” Once the page and the model have loaded, yes: you can blur with the Wi-Fi off.

“Why is the first detection slow?” The 45 MB model downloads once and is cached afterwards; every later photo takes a second or two. “Can I blur only part of the background?” The blur applies to everything outside the subject mask, but you can edit that mask with the AI brush — paint an object into the mask to keep it sharp, or erase to blur it. “Does it work on group photos?” Yes; the model detects people generally, and the brush lets you add or remove individuals from the mask.

The snapshot with the perfect expression and the terrible kitchen behind it, the product on a messy desk, the dog in front of a car park: each is a detection, a slider and a download away from looking like it was shot with a real lens. Keep the blur moderate, fix the edges with the brush where the model stumbled, and judge the result at a normal viewing size. The subject stays pixel-perfect; only the distraction goes soft.

PixGenia’s background blur tool works exactly as described here: the subject is detected on your device by a 45 MB model that is cached after its first load, the blur is rendered in your browser at full resolution, and nothing is uploaded, counted or watermarked — free and unlimited, with no account. Drop in a photo, press Detect the subject, and watch the background melt away.