Soften the background and keep your subject sharp.
A segmentation model traces your subject in the photo, and the blur is applied to everything else, along the real outline, not a shape drawn over the middle of the frame. It deepens towards the top of the frame the way distance does, and the model runs on your own device.
- Blurs along your subject’s real outline
- One dial, re-blurred live
- Nothing uploaded
Drop a photo here, or click to browse
The model runs on your device. Nothing is uploaded.
Separate the subject, then blur the rest.
1. The model finds your subject
The same segmentation model as the background remover runs in your tab and returns your subject cut out on transparency. The first run downloads it; after that it is cached by your browser.
2. The background is patched where the subject was
Removing the subject leaves a hole. Colour from around it is pulled inwards to fill that hole first, so the blur has real background to work with instead of smearing your subject outwards into a ghost of itself.
3. The blur deepens towards the top of the frame
Rather than one flat radius everywhere, the background is blurred in bands that strengthen upwards and crossfade into each other. The ground near your subject’s feet stays the most readable, and the far background at the top goes softest. Then the sharp cutout is composited back over it, its outline feathered by a pixel or so so it does not read as pasted on.
4. It exports as a PNG
The canvas is encoded as a PNG at full resolution and handed to you as a download. No round trip, no queue, nothing kept.
The model comes to your photo.
The photo never leaves the tab. The only thing that travels over the network is the model file itself, downloaded once and then cached by your browser. The picture stays in memory on your own machine.
Because the export is written from a canvas rather than copied from your original file, the result carries no EXIF, no GPS, and no camera information. Blurring a background is also, quietly, a way of stripping the metadata off the photo.
Nothing is stored between runs. Reload the page and the tool has no idea what you were working on.
- Price
- Free, unlimited, no account
- Input formats
- JPEG, PNG, WebP, AVIF, GIF, BMP, and SVG. Vectors are rasterized on the way in
- Subject detection
- Segmentation model, running on your device
- Effect
- Background blurred along the subject’s outline, subject composited back sharp
- Falloff
- Blur strengthens towards the top of the frame, in four crossfaded bands
- Blur strength
- 0 to 30 pixels, at its deepest
- Preview
- Re-rendered as the slider moves, no apply step
- Output
- PNG at the original resolution
The outline is only as good as the photo.
The first run is the slow one
The model has to download before it can look at anything. After that it is cached, and moving the slider only redraws the composite. Your subject is separated once per photo, not once per adjustment.
Under-blur rather than over-blur
Real lenses fall off gradually. Somewhere between 8 and 16 pixels usually looks like a wide aperture, while 30 looks like a filter, which is sometimes what you want, and usually not.
The number is the deepest blur, not the only one
It describes the top of the frame. Lower down the background is progressively sharper, so a setting that looks mild at the horizon is still doing real work behind your subject.
Give the model a subject to find
People, animals, and objects on a surface separate well. A crowd, a landscape, or a subject the same colour as the wall behind it gives the model much less to go on.
Everything else.
Is it free?
Yes. No account, no watermark, no limit. The model runs on your machine, so there is nothing for us to meter.
Does it detect my subject?
Yes. A segmentation model traces your subject and the blur is applied to everything outside that outline. It is the same model the background remover uses.
Do you upload my photo?
No. The model is downloaded to your browser and runs there; the photo itself never goes anywhere.
Why is the first run slow?
The model has to be fetched before it can run. That happens once: your browser caches it, and later photos start straight in on the separating step.
What do I get back?
A PNG at the same resolution as your original. PNG is lossless, so the only thing changed is what the blur changed.
The outline is wrong on my photo. Why?
The model guesses, and it guesses worse when the subject is low-contrast against its background, partly out of frame, or one of several. Try a photo with a clearer subject, or use the background remover to see the cutout on its own.
Does it strip EXIF and GPS?
Yes, as a consequence of exporting from a canvas. The output has no metadata block at all.
Will it work on a phone?
Yes, though the model download and the first separation are noticeably slower than on a laptop, and a very large image costs memory to hold.