The subject is never filtered
The blur is applied to a copy underneath and the untouched sharp subject is drawn over it. Nothing that survives the mask has been softened, which is why the eyes in a portrait stay crisp.
Blur background
Blur the background and keep the subject sharp.
A model on this device separates the subject, then the room behind it is blurred by as much as you choose. The subject itself never goes through a filter.
Runs on this deviceSubject stays sharpFull resolution
The short answer
To blur the background of a photo, add the image, set how much blur you want, and press the button. A matting model runs on your device to find the subject, the original is blurred underneath it, and the sharp subject is drawn back on top at full resolution. The room stays recognisable while the detail that competes with the subject goes away. Nothing is uploaded and the export is the same size as the file you brought.
Blur a background now
Drop a photo, click to browse, or paste one from the clipboard. The first run downloads the model and counts the megabytes as they arrive; after that it is cached and the blur is immediate. The strength is measured against a thousand pixels of width, so the same number looks the same on any size of photo.
Before
Presets
Cut-out quality
Save as
PNG and WebP carry a real alpha channel. JPG does not, so a cut-out saved as JPG lands on a solid colour.
AI model
Checking what this device can run.
Add a photo first.
Checking what this device can do. Nothing is uploaded.
Add a photo and run the tool. The download appears here.
Try another tool on the same photo
The short version
What to know before you start
Blurring a background is the gentlest of the four things you can do with a cut-out, and it is also the most forgiving. A mask that is slightly wrong against pure white shows a halo that anyone can see. The same mask against a blurred version of the same photograph hides its mistakes, because the colours on both sides of the join already match: the blurred pixel next to the subject is made of the pixels that were next to the subject.
What it is not is depth of field. A camera with a fast lens blurs progressively, more with distance, because that is what optics does. This has one photograph and one mask, so everything behind the subject gets the same treatment regardless of how far away it is. On a portrait, where the subject is much closer than everything else, that reads as convincing. On a photo where something sits at the same distance as the subject, it does not, and no setting fixes it.
Strength scales with the picture
The number is a radius per thousand pixels of width, so eighteen looks like eighteen on a phone snap and on a camera file. Absolute pixel blurs are the reason so many sliders feel broken.
Blur hides mask errors, colour does not
The blurred pixels beside the subject are made from the pixels that were beside the subject, so the colours match across the join. This is the most forgiving of the four background tools.
One mask is not depth of field
Real bokeh increases with distance. This blurs everything behind the subject equally, which is convincing on a portrait and wrong on a scene where something sits at the subject's distance.
Less is usually right
Heavy blur makes a subject look pasted on. Around eighteen keeps enough of the room to read as a place, which is what separates a photograph from a composite.
The model downloads once per device
It is cached by the browser and shared with the other background tools here, so the second photo you blur starts immediately and works with the connection off.
Step by step
How do you blur the background of a photo?
Five steps, all inside this page, with the photo read from disk into this tab and nowhere else.
- 1
Add the photo
Drag a file onto the panel, click it to browse, or paste an image from the clipboard. The tool reads the file, applies any camera rotation, and shows its real dimensions.
- 2
Let the model load the first time
The panel counts the megabytes as the model arrives, then spends a few seconds setting it up on your graphics card. That step is silent and it is not stuck. It happens once per device.
- 3
Choose how much blur
Start with the portrait preset at eighteen. Take it down to eight if the room should still read clearly, or up towards fifty for a video call look where the background is a wash.
- 4
Run it
Press the button. The model finds the subject, the original is blurred underneath at the strength you set, and the sharp subject is composited back on top at full size.
- 5
Check the join, then download
Look at where the subject meets the blur, especially around hair and shoulders. Add a pixel of edge softness if the transition is too abrupt, then download the file.
How much blur should you actually use?
Less than instinct suggests. The purpose is to stop the background competing for attention, not to delete it, and a photograph where the room is still legible reads as a photograph. One where it is a coloured wash reads as a composite, and the eye notices the join precisely because there is nothing else to look at.
Around eight keeps the room clearly readable and takes the edge off the clutter. Around eighteen, the portrait preset, is the setting that suits most head-and-shoulders shots and desk photographs. Around thirty-eight the background becomes shapes and colour. Past fifty you are in video call territory, which is a deliberate look rather than a photographic one.
The one case for going heavy is when the background contains something you actively need to obscure, such as a document on a desk or a name on a door. Even then, check the result at full size: a strong blur reduces detail, and it is not a redaction.
Why does a phone's portrait mode look better than this?
Because it knows the distances. A phone with two lenses, or a depth sensor, builds a rough map of how far away every part of the scene is, and blurs in proportion to that distance. The wall behind a shoulder gets a little, the street through the window gets a lot, and the gradient between them is what the eye reads as depth.
A single photograph carries no such map. What this tool has is a subject and everything else, so the blur is uniform behind the mask. On a portrait against a wall, that is close enough to correct that it is difficult to tell. On a garden receding into the distance, the uniform blur flattens the scene and the difference becomes obvious.
Where this wins is that it works on a photograph you already have, from any camera, without needing the depth data that was thrown away when the file was saved. It also works on somebody else’s photograph, which portrait mode cannot do at all.
Should you blur the background or replace it?
Blur when the location is part of the point. A property photograph, a portrait in a workspace, a craft item on a table: in all of those the room says something about the subject and the problem is only that it says it too loudly.
Replace when the location says nothing you want. Product photography is the clear case, and marketplace listings usually insist on it. Use the background changer for a colour or a gradient, or the transparent background tool when the file is going into a layout.
There is also a technical reason to prefer blur when the mask is difficult. A cut-out against blurred pixels of the same photograph is forgiving in a way that a cut-out against pure white is not, so a subject with awkward hair often looks better blurred than replaced.
| Property | Detail |
|---|---|
| Input formats | JPG, PNG, WebP, AVIF, GIF, BMP. HEIC where the browser can decode it. |
| Output formats | JPG, PNG or WebP at the input dimensions. |
| Blur strength | 2 to 80, measured as a radius per thousand pixels of image width. |
| Subject | Never filtered. The sharp original is composited over the blurred copy. |
| Model, detailed | BiRefNet lite, fp16, about 115 MB, MIT licensed. |
| Model, fast | MODNet, fp16, about 13 MB, Apache 2.0 licensed. |
| Download | Once per device, cached by the browser and shared with the other background tools. |
| Backends | WebGPU where available, single-threaded WebAssembly everywhere else. |
| Not depth of field | One uniform blur behind the mask, because a flat photo carries no depth map. |
| Where it runs | In this browser tab, on your device, in a background thread. |
Slide the table sideways to see every column.
Worth avoiding
Where this usually goes wrong
Turning the blur all the way up
A background that is pure wash makes the subject look pasted on, and it draws the eye straight to the join. Eighteen suits most portraits; fifty is a video call look, not a photographic one.
Using blur as redaction
A strong blur reduces detail and does not destroy it, and it is applied to a picture that still exists. If something in the frame must not be readable, crop it out instead.
Expecting progressive depth on a landscape
This blurs everything behind the subject by the same amount. On a scene that recedes into the distance the result flattens rather than deepens, and no setting changes that.
Blurring a photo where something shares the subject's distance
A cup beside a face is at the same distance and will still be blurred, because the mask found the face and not the cup. Check for objects the eye expects to stay sharp.
Blurring after resizing down
Do the blur on the largest version you have. Working on a small copy and enlarging afterwards softens the subject too, which is the one thing this tool is meant to avoid.
Saving to PNG by default
The result is opaque, so the alpha channel has nothing to carry, and PNG makes a photograph several times heavier. JPG at high quality is the right choice unless a design tool is next.
Questions
What people ask about this tool
Every answer is in the page source, so a reader and an answer engine get the same words.
How is this different from a phone's portrait mode?
Portrait mode has depth information. A phone with two cameras, or a lidar sensor, knows roughly how far away every part of the scene is, so it can blur progressively: the wall behind the shoulder a little, the window across the room a lot. This tool has one flat photograph, so it separates subject from background and blurs everything behind by the same amount. On a head-and-shoulders shot the difference is hard to see. On a scene with real depth, a phone's version looks more natural.
What does the blur strength number mean?
Pixels of blur radius per thousand pixels of image width. It is scaled rather than absolute so the same number produces the same visual effect on a small phone snap and on a large camera file. A blur measured in absolute pixels is invisible on a 6000 pixel photo and obliterates an 800 pixel one, which is why sliders that do not scale feel broken on half the images people bring.
Is the subject still full resolution and sharp?
Yes. The blur is applied to a copy of the original and then the untouched sharp subject is drawn over the top, so nothing that survives the mask has been through a filter. The export is the same pixel dimensions as the file that went in, with no watermark.
Why is there a hard line between the sharp part and the blurred part?
Because there is one. Real optical blur increases gradually with distance, and a mask has a single boundary. On a portrait this reads as convincing because the subject really is much closer than the wall. On a photo where the subject stands next to something at the same distance, that something gets blurred too and the result looks wrong. A pixel or two of edge softness makes the join less abrupt.
Can I use this for a video call screenshot or a professional headshot?
Yes, and the video call preset uses a heavy blur for exactly that look. For a headshot, a moderate setting around eighteen keeps enough of the room to read as a place while removing the detail that competes with the face. Going too strong is the most common mistake: it makes the subject look pasted on rather than photographed.
Does this work on a photo with several people?
The model finds the salient subject or subjects rather than one person, so a group standing together usually stays sharp as a group. Someone well behind the main subject may be treated as background and blurred, which is often what you want and occasionally is not. Check the result before using it.
Is the photo uploaded anywhere?
No. The first run downloads the AI model from a content delivery network, and that is the only network traffic the tool causes. After that the model is cached, the photo is read from disk into this tab, and the whole job happens in a background thread in the same tab. It keeps working with the connection off.
Should I blur the background or replace it?
Blur it when the room is part of the story and only the clutter is the problem, which is most portraits, desk shots and property photos. Replace it when the background carries no information you want, which is most product photography and every marketplace listing. Blurring also hides mask errors better, because a soft join against a soft background is much more forgiving than a soft join against pure white.
Keep the room, or lose it entirely.
Replace the background with a colour or a gradient, take a transparent PNG into your own layout, or set the exact size a platform asks for. Every one of them runs on your device.