Video face blur and person path tracking
Finds the people in a video, gives each one a layer and masks them all. Untick the people to leave visible, move drawn masks with keyframes, and keep places like the CCTV clock clear with no-mask zones.
Choose a video. It is processed only on this device.
Your video is not uploaded. The tool only follows the same person within one video; it does not identify anyone from their face.
Press “Start finding” to get one layer per person. Everyone is masked at first.
How to use & FAQ
Is my video sent anywhere?
No. Reading the video, finding people and faces, telling people apart, blurring and making the new video all happen inside your browser. Apart from downloading the model bundle once (about 52 MB), no internet is used.
A small person in the background, or someone standing still, was not masked. What can I do?
The frame is searched in pieces so small people are found too, and a face found once is followed into the frames before and after it (dashed circle). If someone is still missed, choose “Thorough” in Find settings, and turn on anime mode for anime or game video. With “Show marks” on, anyone without a circle can be covered by hand with “Draw a mask”.
Does it work with AVI, WMV and other videos? Can I choose the save format?
Yes. Videos the browser cannot read directly, such as AVI, WMV, FLV and MPG, are turned into an MP4 on this device after a one-time converter download (FFmpeg, about 32 MB). If the video inside uses a common codec such as H.264 this takes seconds; older codecs can take about as long as the video. When saving, choose MP4, MOV, MKV, WebM, TS, AVI, WMV, FLV, MPG, GIF or WebP under “Save format” in “Save settings”.
How does it tell people apart?
It combines three things: (1) where the person should be given their walking direction, speed and size; (2) the colour mix of four body parts — head/hat, top, bottom and shoes; (3) a same-person model trained on surveillance footage that compares clothing, bags, carried items and build all at once. Clearly different items or clothes (backpack vs none, skirt vs trousers) stop a link. Sex and age are guesses from looks and only count a little.
Can a face be missed?
Very small faces or faces turned away can be missed. So the head area of every visible body is covered too, and neighbouring frames are combined so there is no gap. In the preview a solid face circle means a face was found, dashed means an estimate. If it is off, select that person and drag the circle into place.
How is the path made?
Open a person’s layer and turn on “Show path” to draw their foot point (bottom centre of the box) as a line. Hidden stretches are dashed and marked “lost” in the table. Positions are screen coordinates, so a fixed camera (CCTV) works best.
What if it follows the wrong person?
If one person ended up in two layers, merge them with “Merge with…”. If two people got mixed into one layer, move to where it starts and press “Split here”.
Which videos work?
MP4, WebM and MOV files your browser can read. The result is saved as MP4 (H.264), or WebM if your device cannot encode H.264. Long high-resolution videos use a lot of memory, so a computer works better than a phone.
How do I keep the CCTV clock from being masked?
Press “No-mask zone” and drag a box over the clock. No mask of any kind — people or drawn — is drawn inside that box.
Saved AI files
Saved AI files
| Model | Used for | Size | Status | Location | Device load | Manage |
|---|
Location
AI files are stored in your browser. Connecting a save folder moves them into its models folder, where new downloads are also saved. You can export or delete them here. Disconnecting a folder leaves its existing files in place.
Websites cannot open Explorer or read a folder’s full path. Enter the connected folder’s full path once to copy its models path.