A phone’s portrait mode isn’t always on when the good shot happens. Add a photo or a video and the background behind every person turns soft, with sliders for blur strength, edge feather and background colour. Nothing is uploaded.
Processing locally on your device
Overview
Optical background blur comes from a wide aperture: anything outside the plane of focus spreads into soft discs of light. Portrait modes fake this in two steps. First they work out which pixels belong to people, then they blur everything else. Almost all of the realism comes from the first step. A well-placed edge around hair and shoulders looks natural, while a sloppy one creates the halo that gives software blur away.
This tool finds people with Google’s MediaPipe Image Segmenter. For photos it starts with the Quality model, Selfie Multiclass (about 16 MB), which labels hair, face, body and clothes separately from the background. For video it starts with the Fast model, Selfie Segmenter (about 0.25 MB), which is quick enough for a live preview while the clip plays. You can switch either way, and both are Apache-2.0 licensed. The models run through MediaPipe’s WebGL delegate on your graphics card, or on the CPU with WebAssembly if WebGL isn’t available. They are cached after the first download, together with the roughly 12 MB runtime.
The blur is a Gaussian blur rendered at a quarter or half of the resolution and scaled back up. That is faster than blurring at full size and looks almost identical, because a blurred image has no fine detail to lose. The person is cut out with the upscaled mask and placed back on top, so they stay at full resolution. A desaturation slider can fade the background towards black and white for a stronger focus effect. Photos are exported at their original resolution as JPG or lossless PNG.
Videos are processed frame by frame. A hidden copy of the clip is stepped through, each frame is segmented and blurred, and FFmpeg (WebAssembly) encodes the result to H.264 MP4 with the original audio, at 480p, 720p or 1080p. Temporal smoothing blends masks between frames so the edges don’t shimmer. Expect a 30-second 720p clip to take a few minutes on a laptop. Short clips are the sweet spot.
| Scene | Blur | Feather | Desaturate |
|---|---|---|---|
| Headshot, plain room | 40–60% | 1–2 px | 0% |
| Busy street or crowd | 60–80% | 2–3 px | 20–40% |
| Webcam video | 40–60% | 2 px | 0–20% |
| Hiding documents or screens | 70–100% | 2 px | 0% |
Drop in a photo (JPG, PNG, WebP, HEIC) or a video (MP4, MOV, WebM) with people in it.
Let the segmentation model load. Photos are blurred automatically, and videos show a live preview next to the original.
Set the blur strength and, optionally, desaturate the background.
Adjust the mask threshold and edge feather until the outline looks natural.
Download the photo as JPG or PNG, or pick an export range and resolution and export the video as MP4.
Use cases
Turn a picture taken in front of a cluttered office or a busy street into a clean LinkedIn or team-page portrait where the background melts away.
Give a webcam or phone recording a shallow-depth look so the speaker stands out, without re-shooting with a fast lens.
Soften documents, screens, family photos or a whiteboard in the background of a photo or a recorded update, while still showing a real room.
Make friends and family pop out of crowded wedding, conference or landmark shots without any manual masking.
Pro tips
Answers to common questions about the portrait blur.
It keeps every detected person sharp and blurs the rest. Objects people hold, pets and furniture count as background, so they are blurred too.
The models find people at 256 × 256 pixels internally, and the mask is scaled up to your photo’s resolution. Strands of hair against a busy background can blend into the blur. The Quality model and a slightly looser mask help.
No. It uses a person mask, not a depth map, so the whole background gets the same amount of blur. That is how most phone portrait modes looked before depth sensors, and it works well for people in the foreground.
Each frame is segmented, blurred and composited in your browser at roughly 5–20 frames per second, then encoded to MP4. A 30-second 720p clip at 30 fps (900 frames) usually takes a few minutes. 480p is faster.
No. The output keeps the full original resolution. Choose PNG for a lossless file or JPG (quality 92) for a smaller one.
No. ToolEnhancer processes files on your own device inside the browser. They are never sent to, stored on or seen by our servers.
Yes. The portrait blur is free to use with no sign-up, no watermarks and no daily limits.