How it works
How to blur your background on camera without a green screen
Background blur in tools like Zoom, Google Meet, or a browser-based recorder doesn't use a green screen or any special hardware — it uses a machine learning model that looks at each video frame and estimates, pixel by pixel, which parts are "you" and which parts are "background."
The basic idea
The model is trained on a large number of images of people to recognize the rough shape of a person — head, shoulders, torso — against all kinds of backgrounds. For each frame of live video, it outputs a mask: a grayscale image where white means "probably a person" and black means "probably background." The app then blurs everything the mask marks as background and leaves the person layer sharp on top.
Why it's not always perfect
Fast movement is the main weak point. A quick turn of the head or a wave can genuinely blur the camera frame for an instant, and a model trained mostly on sharp images doesn't classify a blurry frame as confidently. This shows up as a brief gap or fringe around fast-moving edges — hair and hands are usually the first place it happens.
Lighting matters more than most people expect, too. A well-lit face lets the camera use a faster shutter speed, which reduces motion blur at the source before it ever reaches the segmentation model. A dim room forces a slower shutter, and slower shutters blur more during any movement.
Live blur vs. processed-after-recording blur
There are two different approaches, with a real trade-off between them. Applying blur live, while you're recording or on a call, has to make a decision every frame in real time — it can't look ahead. Applying it afterward, as a processing step once the recording is finished, can look at frames both before and after any given moment, which produces a noticeably steadier result because a single bad frame gets smoothed out by the ones around it instead of showing up as a visible glitch.
The trade-off is time: processing after recording takes a little while once you stop, rather than being instant. For a short recording — the kind used for a job application, a proposal, or a resume clip — that wait is usually a few seconds to a couple of minutes, which is a reasonable exchange for a cleaner result.
Simple things that improve the result
- Face a light source rather than having it behind you.
- Keep movements slightly slower and more deliberate than feels natural.
- Plain, high-contrast clothing against your actual background helps the model separate the two more cleanly than clothing that blends in.
Jikoreel uses this processed-after-recording approach specifically to get a steadier result than a live call would give you — worth knowing if you've tried background blur elsewhere and found it glitchy during movement.
Try it yourself — the first minute is free.