How to blur bystanders in vlog and street footage
- Step 1Load your clip (Pro + Media) — Face blur is a Pro + Media feature. Drop a single MP4, MOV, MKV, WebM, AVI, M4V, or TS clip (camera files up to 100 GB). The page reads duration/resolution locally — your footage never leaves the browser.
- Step 2Pick a sample rate for foot traffic — A quiet street is fine at the default 4 Hz. A busy sidewalk where people cross quickly benefits from 8–10 Hz so brief passers-by are caught and the blur follows movement tightly. Range is 1–15; higher means a longer detection stage.
- Step 3Set blur strength to taste — For a natural look that doesn't scream 'censored', strength 20–25 softens faces like background depth-of-field. For clearer anonymisation, push toward 35–40. Range 5–50, default 25; the radius is auto-clamped so distant faces aren't over-processed.
- Step 4Tune padding for walking subjects — Default padding 0.25 is usually enough. If a bystander's hair or shoulder-up profile peeks past the blur as they move, raise to 0.35. Padding scales with each face's size, so it adapts to near vs far automatically.
- Step 5Run it and grab a coffee — You'll see a one-time detector warm-up, then the sample loop, then 'Blurring faces · N faces'. Output is H.264 MP4 with your audio intact — ready for your editor.
- Step 6Quick-scrub for missed faces — Skim the result, especially shots with distant figures or quick side-on passes (the detector's weak spots). For the occasional miss, draw a fast rectangle with video-redactor, or swap to the mosaic look in face-pixelate.
Vlog scene presets (manual values)
Starting points for common vlog scenarios, within the tool's real ranges.
| Scene | Sample rate | Strength | Padding |
|---|---|---|---|
| Quiet street, you walking and talking | 4 Hz | 25 | 0.25 |
| Busy sidewalk / market with crossing people | 8–10 Hz | 30 | 0.35 |
| Cafe / indoor with seated background people | 4–6 Hz | 25 | 0.3 |
| Subtle, natural look (you want soft, not censored) | 4 Hz | 20 | 0.25 |
What it does to your clip
Quick facts for fitting this into an edit.
| Aspect | Behaviour |
|---|---|
| Your face | Also blurred if detected — there's no exclude/allow-list |
| Audio / narration | Copied unchanged (-c:a copy) |
| Output | H.264 MP4, CRF 20, +faststart |
| Max faces blurred | 12 most-detected clusters per clip |
| Distant background faces | May not register (short-range model) |
Tier access
Pro + Media required (not streamable).
| Tier | Access | Max file | Batch |
|---|---|---|---|
| Free / Pro | Blocked (upgrade) | — | — |
| Pro + Media | Full | 100 GB | 50 |
| Developer | Full | 100 GB | Unlimited |
Cookbook
Fast workflows for creators who shoot in public and post on a schedule. Tuned for the vlog audience, using only the real controls.
Walk-and-talk down a city street
You're the subject, strangers cross behind you. Note your own face gets blurred too if detected — film your pieces-to-camera separately if you need your face visible.
Input: street-walk.mp4 (4K, 3m)
Options: sampleHz 8 · strength 30 · padding 0.35
Result: background passers-by blurred; your face also
blurred when on camera (no allow-list).
Tip: shoot your direct-to-camera bits in a private
spot, blur only the street b-roll.Busy market — raise the sample rate
Lots of fast movement and crossing faces. Higher Hz catches brief appearances; expect the twelve-track cap in dense crowds.
Options: sampleHz 12 · strength 35 · padding 0.35 Console may show: capped to 12 most-detected Meaning: the 12 most-seen faces blur; a couple of transient background faces might not. Fix: quick video-redactor pass on any you spot.
Subtle, natural blur for a travel vlog
You want the look to feel like shallow depth of field, not a redaction. Low strength does that.
Options: sampleHz 4 · strength 18 · padding 0.25 Reads as soft background rather than 'censored'. Still anonymises casual recognition for most viewers.
Keep your narration perfect
Audio is copied, so blurring doesn't degrade your voiceover or ambient sound at all.
face-blur -> video re-encoded (faces hidden)
-> audio identical to source
Drop straight into your NLE; sync stays intact.Strip location before you post
Your phone may have stamped GPS into the clip. Blurring faces doesn't remove that — chain the scrubber.
face-blur (strangers hidden) -> metadata-scrubber (GPS, device, date removed) -> upload Now neither the faces nor your filming location leak.
Edge cases and what actually happens
Your own face gets blurred
By designThere's no allow-list — every detected face is blurred, including yours. If you need your face visible, shoot direct-to-camera segments separately and only run face-blur on the public-facing b-roll, or use video-redactor to manually redact just the strangers.
Distant figure at the end of the street not blurred
Expected limitationThe short-range model is tuned for near-camera faces, so a small/distant figure may not register. Raise the sample rate to catch a clearer frame, or accept it if they're genuinely unidentifiable at that distance; otherwise patch with a manual rectangle.
Quick side-on pass slips through
Expected limitationA pure side profile crossing fast can fall between samples or go undetected. Raise the sample rate (up to 15 Hz) to shrink the gap, and review the export.
Crowd exceeds twelve tracked faces
By designOnly the twelve most-detected face clusters are blurred. A packed market may leave a few low-hit background faces visible. Raise the sample rate so more faces accumulate hits, then quick-redact the rest with video-redactor.
No faces detected
ErrorIf a shot has no clear faces (all very distant or extreme angles), you'll see 'No faces detected. Try lowering the sample rate or use a clearer source.' If there genuinely are faces but they're tiny, manual redaction is the fallback.
Clip comes out as MP4 not MOV
By designOutput is always H.264 MP4 — there's no format choice. Most NLEs and platforms take MP4 fine. If your edit needs another container, run video-transcoder.
Blur looks too strong / too obvious
ExpectedLower the strength toward 18–20 for a softer, depth-of-field feel that anonymises casual recognition without screaming 'censored'. Re-run with the lower value.
Long 4K clip takes a while
ExpectedDetection and the re-encode run in your browser and scale with resolution and length. A long 4K street session takes real time; lower the sample rate to speed detection if you're in a hurry.
Frequently asked questions
Will it blur my own face too?
Yes, if your face is detected — there's no allow-list to keep one person visible. For pieces-to-camera, shoot them separately and only blur the b-roll, or use video-redactor to manually redact just the strangers.
Does my raw footage upload anywhere?
No. Detection (TensorFlow.js) and blur (FFmpeg.wasm) both run in your browser. Your footage, your face, and your location stay on your machine. Only an anonymous run counter is logged.
Will it follow people as they walk through frame?
Yes — each face's per-sample boxes are clustered into a union region that covers their whole path while they're on screen. Someone walking across stays blurred; raise the sample rate if a fast mover briefly slips.
Can it keep up with a weekly upload schedule?
It's a single FFmpeg pass with the fast ultrafast x264 preset, plus a detection stage. A few-minute clip is quick; long 4K sessions take longer. Batching (Pro + Media: 50 files) lets you queue several clips at once.
How strong should the blur be for a natural look?
Around 18–25 reads like background depth-of-field while still anonymising casual recognition. Push to 35–40 if you want clearer redaction. Range is 5–50.
What if it misses a face in the background?
Raise the sample rate to catch more frames, then review and patch any miss with a quick rectangle in video-redactor. Always skim the result before posting.
Does the audio change?
No — it's stream-copied (-c:a copy). Your narration and ambient sound are byte-identical to the recording, so sync and quality are preserved.
What format do I get for my edit?
H.264 MP4 with +faststart, which every major NLE and platform accepts. For a different container, follow with video-transcoder.
How do I also remove my filming location from the file?
Run metadata-scrubber after blurring to strip GPS coordinates, camera info, and capture date. The blur handles faces; the scrubber handles the file's hidden data.
Which tier do I need?
Pro + Media (£19/month) or higher. The tool isn't streamable and needs full-frame access, so Free and Pro show an upgrade prompt.
Can I get a mosaic look instead of a blur?
Yes — face-pixelate uses the same detection but applies a mosaic/pixelate, which reads as a more deliberate TV-news style. Same controls plus a pixel-size setting.
What's the max file size?
100 GB per file on Pro + Media (and Developer), which comfortably covers long 4K camera files. Larger or longer files just take more time to process locally.
Privacy first
Every JAD Video tool runs entirely in your browser via WebCodecs and FFmpeg (WebAssembly). Your video files never leave your device — verified by zero outbound network requests during processing.