Give this skill to your agent

Copy the prompt and paste it into Claude Code, Codex or any agent that can read a web page. It installs the skill and tells you what it does.

Install this skill for me.

Read https://cwyz.com/skills/video-face-anonymizer-en
The raw SKILL.md is at https://cwyz.com/api/skills/video-face-anonymizer-en/raw

Save it as video-face-anonymizer/SKILL.md in my skills folder (Claude Code: ~/.claude/skills/ for every project, or .claude/skills/ for this project only).

Then tell me in two sentences what it does and when it applies. Do not run anything it describes until I say so.

Download .zip View SKILL.md

Or unzip the download into ~/.claude/skills/ (every project) or .claude/skills/ in your project. A skill tells an agent what to do. Read it before you install it, especially one you did not write.

Video Face Anonymizer Blurs Every Face Automatically

Automatically detects, tracks and obscures faces in video, with mosaic, blur or solid black masks, preserving the original motion and audio. Processes locally and outputs Telegram-compatible H.264 MP4. SCRFD suits small distant faces in groups; YuNet is faster.

Use caseAutomatically blur faces in video (mosaic, blur or solid black), keep the audio, process locally.

Machine translation. Read the original

Video Face Blurring

Automatically detects, tracks and obscures faces in video, keeping the original motion and sound, and outputs an MP4 compatible with Telegram and common players. Everything runs locally; no video is uploaded.

What It Does

Automatically detects and tracks faces in the frame, masking them frame by frame.
Three mask types: mosaic (default), blur, and solid black mask (strongest privacy protection).
Preserves the original video's motion and audio.
Outputs H.264 + yuv420p for good compatibility.

Two Detection Modes

SCRFD (default): suited to distant, small, profile, low-light and multi-person frames; more accurate, but slower on CPU.
YuNet: suited to well-lit close-up or medium shots; faster.

Installation

pip install opencv-python numpy insightface onnxruntime

You also need to install ffmpeg. The core of this skill is the author-provided script scripts/face_mosaic.py; run it as is, and do not write your own version.

Basic Usage

python scripts/face_mosaic.py input.mp4 output.mp4

Common Options

--mode mosaic: mosaic (default)
--mode blur: blur
--mode black: solid black mask, strongest privacy protection
--detector yunet: switch to fast detection
--det-size 1600: improve detection of small distant faces
--score: detection confidence threshold
--margin: enlarge the mask area
--hold: keep the mask in place during brief missed detections

Notes

The first use of SCRFD automatically downloads the InsightFace buffalo_l model, about 275 MB.
Very small, heavily blurred or fully occluded faces may still go undetected.
For high-privacy scenarios, use --mode black: mosaic and blur are only visual obscuring, not irreversible anonymization. Leave 20%–35% margin around the mask boxes.
Always spot-check before delivery: at the beginning, middle and end, and in profile, varying-distance, occluded and transition shots, check at least six time points. If any recognizable face remains, redo it.

Source

This skill was packaged by the user. The author's script uses open-source components: InsightFace SCRFD / buffalo_l for detection, OpenCV Zoo's YuNet for fast detection, OpenCV for frame processing, and FFmpeg/libx264 for encoding. Please verify the blurred output yourself; the tool does not guarantee that it catches 100% of faces.

Done with this one? Mark it to track your path.Sign in

Save it for laterSign in

Questions about this piece?Ask AI about this