Skip to content

Blur every face, plate and name tag it finds, and see each frame it covers.

Video Blur AI numbers everything it finds, blurs it by default, and shows the frames where a mask is missing, before the file leaves your computer.

Assumed means the engine must report which frames have a box; otherwise it cannot show them.

Supervised by NHK Technologies

Design proposal · not the official AVCLabs site

Skip the scene
Stock footage of a conference audience, numbered in order of first appearance: Face 3, Face 9, Face 13, Name tag 14. Face 9 is kept clear; the rest are in Mosaic.
Computed offlineHand-authoredMediaPipe found the faces. 13's badge is blank.

How it works

Design proposal

  1. Find

    A scan numbers every face, plate and name tag by first appearance.

  2. Decide

    Everything starts blurred. Choose who stays clear, by number; it shows on the frame, in the list and on its lane.

  3. Check

    Each subject has a lane, and frames missing a mask show in red. Assumed Review lists every gap, longest first, since a long gap shows a face longest.

  4. Export

    Before the file is written, a summary lists what is blurred, kept clear and unchecked; a gap may need no mask, so it never stops export. Assumed

Each subject’s masks join into a tube through time; red marks the frames missing a mask.

  1. 00:00:14:20
  2. 00:00:15:01
  3. 00:00:15:07
  • 3, Face, Blur
  • 9, Face, Keep clear
  • 13, Face, Blur, 11 frames with no mask from 00:00:14:21 to 00:00:15:06
    13 GapFace · GapAssumed

    00:00:14:21 to 00:00:15:06 · 11 frames

  • 14, Name tag, Blur

Computed offlineHand-authoredFace tracks by MediaPipe on a stock clip, the badge by hand; drawn for this page.

Users report unblurred frames (Trustpilot, 2024, 2025).

For broadcast

Supervised by NHK Technologies for edit rooms with no internet.

PR TIMES, Medianavi · Nov 12, 2024 Nov 18, 2025

  • Keep the host clear

    Mark the presenter; every other face it finds stays blurred.

  • Mask for finishing

    Export the mask as a PNG sequence for your editing software.

  • Offline edition

    For rooms with no internet (Japan).

What it finds

  • Faces

    A face lost and found again may get a new number, blurred, with no red between unless linked.

  • Plates

    License plates.

  • Name tags

    Badges and tags with names on them.

  • Keyword

    Anything you can name: type "logo", "screen", "bottle".

  • Regions

    A fixed area, drawn once.

  • Frame by frame

    Paint a mask on the frames that need it.

Treatments

After: MosaicBefore
Computed offlineHand-authored

Drag to compare.

Try it on a clip

Pick a stock clip, click a face to keep it clear, and see its gaps.

Runs Google MediaPipe in your browser, not Video Blur AI's models: it shows the workflow, not their accuracy. Nothing leaves this page.

The demo needs JavaScript and WebGL2 in your browser.

Specs

From the user guide, as of 2026-10-18

Platforms
Windows 11 and 10, 64-bit · macOS 12 or later, Apple silicon only
Memory
8 GB minimum, 16 GB recommended
Graphics
  • NVIDIA GTX 10 series or later
  • Intel and AMD through OpenCL
  • Keyword tracking needs a GTX 16 or RTX card with 8 GB or more
  • Keyword tracking on Apple silicon: not yet confirmed

Pricing

Windows prices in USD. Each license covers one computer.

As listed on avclabs.com, 2026-10-18.

Japan, according to Medianavi (Nov 2025): download edition ¥30,000 per PC, tax included; offline edition quoted individually.

FAQ

A face or tag the scan never found is on no list. Play the file once at 2× before export; paint a mask where you see one. At 2×, a one-frame face can slip by.

No. Detection and rendering run on your computer.

Yes. Choose Keep clear on that person's number. Their name tag stays blurred. Where boxes overlap, blur wins.

No. AVCLabs lists no feature that edits sound, so a name said aloud stays in the file.