How to blur a face or a licence plate in a video

A bystander walked through your shot. A licence plate is legible in the establishing wide. Your screen recording has a client's name in a browser tab for four seconds. Each of these is a small job, and each has a different correct answer.

Four tools, and they are not interchangeable

Blur a region. Effects → Blur softens either the whole frame or a rectangle or ellipse you drag over the thing. This is the workhorse for faces and plates.

Mask. Crops the clip to a rounded rectangle, ellipse or circle with a feathered edge. It hides everything outside a shape rather than obscuring something inside one — useful when the sensitive thing is at the edge of frame and you would rather lose the edge.

Spotlight. Dims the whole frame except a highlight you drag into place. Written as an attention tool, it doubles as a privacy tool for screen recordings: instead of blurring six things, dim everything except the one region you meant to show.

The AI eraser. Removes the thing entirely and reconstructs what should be behind it. Different job, different price, and the one case where movement is handled for you.

The fact that decides the technique

A blur region stays where you put it for the clip's duration. It does not track a moving subject.

That is not a gap to work around later — it is the whole method. The technique is:

1. Watch the clip and find the moments where the subject moves enough to leave the region.
2. Split the clip at each of those moments.
3. Blur each piece with its own region, placed for that piece.

A person walking across a five-second shot might need four splits. It takes three minutes and it is reliable, which the alternative is not. Blurring an oversized region that covers the whole path is the mistake here: it obscures half the frame and draws far more attention than the face did.

If the subject moves continuously and unpredictably — someone walking towards camera through a whole shot — stop splitting and use the eraser instead. It handles movement, which a painted region cannot, and it is the right tool the moment the split count passes about six.

Sizing it so it actually covers

The three failure modes, in order of how often they happen:

Too small. A region sized to the face at the moment you placed it will expose an ear, a jawline or a chin as the head turns. Faces are recognisable from very little. Go bigger than looks necessary — head and hair, not features.

Too weak. A light blur on a licence plate leaves the character shapes intact, and character shapes are what a plate is. Text needs a strength where you cannot tell how many characters there are, let alone which. Check by pausing and looking at the edges of the blurred region, where the falloff is weakest.

Only in one place. The plate is legible in the wide, the cutaway and the reflection in the window. Watch the whole edit for the thing you are hiding before you decide you are done — the reflection is the one everyone misses.

What blurring hides, and what it does not

For an accidental bystander or a plate in the background, a strong blur is proportionate and fine.

For anything genuinely sensitive — an API key on screen, a document, a real address — do not rely on softening. A blur is a filter over the original pixels, and low-strength blur over high-contrast text preserves more of the underlying structure than people assume. Use something that replaces the pixels rather than smearing them: a solid shape, a mask that crops the region out of frame, or the eraser.

The simplest version of this: if it can be cropped out, crop it out. A reframe or crop that loses the bottom eighth of the picture removes the taskbar with the client's name in it permanently, and there is nothing left to reverse.

Two more things blur does not cover:

• The audio. You blurred a face and left the person's name spoken twice in the narration. Cut the audio or re-record the line, and re-caption afterwards.
• The captions. If captions were generated before you fixed the audio, the name is still on screen in text. Regenerate them.

Screen recordings are a different shape of problem

In a screen recording the sensitive material is usually text, static, and there is a lot of it: notification banners, browser tabs, an email address in a header, a sidebar of real customer names.

Blurring each of those individually is a bad trade. Better options, roughly in order:

• Crop the frame so the chrome, the tabs and the sidebar never appear.
• Zoom in on the region that matters with a Ken Burns move — it removes the surroundings and makes the UI legible on a phone at the same time.
• Spotlight the area you are demonstrating, so everything else dims. It reads as deliberate direction rather than as censorship.

And the version that saves the most time: record the demo in a clean profile with no real data in it. Every minute spent on that is five minutes not spent blurring.

Check the export, not the preview

Blur, masks and the spotlight apply to the composited layer and survive transitions, so what you see is what renders. Still, watch the finished file once, at full size, and step through the frames either side of each split — the join is where a region that was placed a frame late shows a clean face for a single frame.

One frame is enough. It is the frame that gets screenshotted.

The order that works

1. Watch the whole edit and list every appearance of the thing, including reflections and cutaways.
2. Crop or reframe it out wherever that is possible — the only fully reliable fix.
3. Split the clip at each point the subject moves.
4. Blur each piece with its own oversized region, at a strength that destroys structure.
5. Switch to the eraser when the splits pass about six, or the subject never stops moving.
6. Fix the audio and the captions, which blur does not touch.
7. Step through the exported file at every split.