How to mosaic blur faces before posting to social media
- Step 1Export the clip from your phone — AirDrop or transfer the MP4/MOV to the device with the browser, or open the JAD tool on your phone's browser directly. Nothing uploads — the file is read locally. Face pixelate needs the Pro-Media tier; it is not on Free or Pro.
- Step 2Drop it onto the tool — Drag the clip onto the drop zone. The first run warms up the detector once (
Loading face detector) while TensorFlow.js compiles WebGPU shaders and loads the MediaPipe model. - Step 3Keep the sample rate at 4 for normal motion, raise it for fast pans —
Sample rate (Hz)is 4 by default. For a quick whip-pan past a crowd, raise it toward 8–12 so faces that flash by for a fraction of a second are still caught and tracked. - Step 4Pick a pixel size that hides the face at feed resolution —
Pixel size(4–40, default 16). On vertical phone video, faces are often large — use 20–28 so identity is hidden even after the platform re-compresses your upload. Smaller blocks can survive less well through TikTok/Instagram transcode. - Step 5Add padding so hair and chin are covered —
Padding (0–1), default 0.25, grows each box outward. Bump to 0.3–0.4 if the mosaic looks like it's leaving a recognisable jawline or hairline at the edges. - Step 6Run, preview, and post — The bar shows detection then
Pixelating faces · N faces. Download the MP4, scrub through it to confirm every stranger is covered, then upload to your platform. Remember it mosaics every detected face, including yours.
Settings for common social clips
Starting points by clip type. All three controls live in the Face pixelate (AI) panel; there are no presets in the UI, so these are manual recommendations.
| Clip type | Sample rate | Pixel size | Padding |
|---|---|---|---|
| Static vlog / talking to camera with people behind | 4 | 18 | 0.25 |
| Walking street clip, faces passing | 8–10 | 20 | 0.3 |
| Kids' party / many close faces | 6 | 24 | 0.3 |
| Fast pan across a crowd | 10–12 | 20 | 0.3 |
| Close-up of one stranger to hide | 4 | 30 | 0.35 |
What the tool does and does not do
Set expectations before you post.
| Question | Answer |
|---|---|
| Can I pixelate only the strangers, not me? | No — it mosaics every detected face. Crop yourself out, or use the video redactor for manual region control |
| Output format? | Always MP4 (H.264, yuv420p, +faststart) — feed-ready for all major platforms |
| Is my trending audio kept? | Yes — audio is stream-copied, byte-identical and in sync |
| Does it remove my location tag? | No — only pixels change. Run metadata scrubber to strip GPS |
| Will it work on my phone browser? | Yes if your phone browser supports WebGPU/WebGL; falls back to CPU (slower) otherwise |
Cookbook
Phone-footage scenarios with the settings that work, written as before → after notes.
Café vlog — strangers in the background
You're talking to camera; two strangers sit behind you. The detector will mosaic them and you. If you want only them hidden, frame so you're not detected or accept being mosaiced too.
Input: cafe_vlog.mov (you + 2 background faces)
Settings: Sample rate 4, Pixel size 18, Padding 0.25
Result: 3 faces mosaiced (incl. yours), audio = original
Tip: to keep your own face, crop yourself out is not possible
mid-frame — record so only backgrounds need hidingKid's birthday — other parents' children in shot
Lots of faces close together. Raise pixel size for strong anonymisation and padding so hairlines are covered. Processing stays on-device, which is the point when minors are involved.
Input: party_clip.mp4 (5+ faces)
Settings: Sample rate 6, Pixel size 24, Padding 0.3
Result: faces mosaiced as heavy blocks
Note: if >12 distinct tracks form, only the 12 most-detected
are covered — re-check the crowd in the previewWhip-pan past a crowd
A fast pan gives each face only a few frames. Raise the sample rate so they cluster into stable tracks instead of being missed.
Input: market_pan.mp4 (fast horizontal pan)
Settings: Sample rate 12, Pixel size 20, Padding 0.3
Why: 4 Hz may catch each passing face once → no track
12 Hz catches several → reliable mosaic
Trade-off: detection takes longerHide a face AND strip the location before posting
Pixelating doesn't remove the GPS/location metadata your phone embedded. Do both: face pixelate, then metadata scrubber.
Step 1: face-pixelate → faces mosaiced Step 2: /video-tools/metadata-scrubber → strip GPS + device Result: no faces, no location tag — safe to post
A face that didn't get mosaiced
Someone in deep profile or far in the background may be missed by the short-range model. There's no manual picker here — cover it by hand with the redactor.
Symptom: one background face still visible after the pass
Fix: /video-tools/video-redactor → draw a box over it
(then post the redactor's output)Edge cases and what actually happens
It mosaics your face too
By designThe tool pixelates every detected face — there's no 'everyone except me' option. If you must keep your own face visible, frame the shot so only the people you want hidden are in it, or use the video redactor to manually cover specific people instead of auto-detecting all.
Platform re-compression weakens a fine mosaic
Use bigger blocksInstagram/TikTok re-encode your upload. A very fine mosaic (Pixel size 4–8) can look slightly recovered after their transcode. Use 20–28 so the blocks survive compression and identity stays hidden in the published version.
Output is MP4 even from a MOV
By designiPhone MOV files come out as MP4 (H.264). That's exactly what every social platform wants, so no extra step is needed — but if you specifically need MOV back, transcode with the video transcoder.
More than 12 faces at a busy event
Capped at 12 tracksOnly the 12 most-detected tracks are mosaiced. In a packed party or street scene, faces seen in few frames may slip through. Preview carefully, raise the sample rate, and patch any misses with the redactor before posting.
Background faces too small to detect
May be missedThe MediaPipe short-range model targets near-camera faces. Tiny faces deep in the background may not be detected, so they won't be mosaiced. For distant crowds see the street-footage spoke.
Trending audio stays in the clip
PreservedAudio is stream-copied, so a trending sound or your voiceover is preserved exactly. If you need to mute a spoken name or remove copyrighted music, that's a separate step — audio region muter.
Location metadata still present after pixelating
Not removedPixelation only edits pixels in face regions. Your phone's embedded GPS/location and device info are untouched. Run the metadata scrubber before posting if you don't want your home or filming location leaking.
Tool gated on Free / Pro
Tier requiredFace pixelate needs Pro-Media or higher. On Free (1 GB / 1 file) and Pro (10 GB / 5 files) it's locked. Pro-Media is 100 GB / 50 files; Developer is 100 GB / unlimited. There's no length cap, only file size and batch count.
Frequently asked questions
Can I pixelate only the strangers and keep my own face?
No. The tool auto-detects and mosaics every face it finds; there is no 'exclude me' toggle. Either frame the shot so only the people you want hidden are present, or switch to the video redactor where you draw the regions yourself and can cover specific people only.
Will the mosaic survive Instagram/TikTok compression?
Yes if you use a large enough pixel size. Platforms re-encode uploads, which can slightly soften a very fine mosaic. Set Pixel size to 20–28 for vertical phone footage so the blocks remain obvious and the face stays unrecognisable after the platform's transcode.
Is mosaic better than blur for social posts?
It's a style choice. Mosaic blocks read as deliberate anonymisation, which looks intentional and respectful in a feed. A soft blur can look like an accidental focus issue. If you prefer the soft look, use the face blur tool instead — same auto-detection, Gaussian-style blur.
Does my clip get uploaded to a server?
No. Detection runs in an offscreen <video> element and encoding runs in FFmpeg.wasm, both inside your browser. The file never leaves your device. This matters especially when other people's children are in frame — nothing about the footage is transmitted.
What about the kids' faces — is this safe to use for that?
The processing is fully local, so the footage of minors is never uploaded anywhere. Use a larger Pixel size (24–40) and Padding (0.3+) for strong coverage, and always preview the result frame-by-frame, because the detector can miss small or profile faces — patch any misses with the video redactor.
Does it keep my trending sound / voiceover?
Yes. Audio is stream-copied (-c:a copy), so the soundtrack is byte-identical and stays in sync with the video. If you also need to bleep a name or remove music, do that separately with the audio region muter.
What format comes out?
An MP4 (H.264, yuv420p, +faststart) — exactly the format Instagram, TikTok, Reels, and Shorts expect. The original container (e.g. iPhone MOV) is not preserved; if you need it back, use the video transcoder.
How do I set the pixel size for phone video?
On vertical phone footage faces tend to be large, so use Pixel size 20–28 for a strong, compression-resistant mosaic. The control range is 4–40. Smaller values leave more facial structure and may not survive the platform's re-encode.
It missed a face in the background — what now?
The short-range model can miss small, distant, or profile faces. First raise Sample rate (Hz). If a specific face is still uncovered, draw a manual mosaic over it with the video redactor and post that output. The two passes compose cleanly because both export standard MP4.
Does pixelating also hide where I filmed?
No. Pixelation changes only the face pixels; your phone's embedded GPS/location and device metadata remain. Run the metadata scrubber before posting to strip location, camera model, and capture date.
Can I do this on my phone?
Yes, in a phone browser that supports WebGPU or WebGL. The detector falls back to CPU if neither is available, which works but is slower. The whole pipeline (detect + encode) runs in the browser tab, so it can be done entirely on the phone.
Which plan is required?
Pro-Media or higher. Face pixelate is gated on Free and Pro. Pro-Media gives 100 GB per file and 50 files per batch; Developer gives 100 GB and unlimited files. Limits are file-size and batch-count, not clip length.
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.