Blurry

Cover faces and strip metadata from photos and videos,
on your own computer, with no network.

⬇ Download for Mac  ·  Windows and Linux (with Python)
_{Version 0.1.0 · Mac with Apple Silicon, macOS 14+ · free and open source (MIT)
Italiano · Threat model · Security}

Blurry finds faces, covers them with a solid black box (or pixelation), and writes a new file with no metadata: no GPS position, no camera model, no dates, no hidden thumbnail of the original. It runs entirely offline. It has no server, no account, no telemetry and no update check, and it refuses to open network connections even if something tries.

Always check the result before you share it: the detector can miss faces, and Blurry does not hide bodies, voices or places. See What it does not do.

Install

Mac

Apple Silicon (M1 or later), macOS 14 or later.

1. Download Blurry.dmg from the Releases page.
2. Open it and drag Blurry onto the Applications folder.
3. The first time, macOS warns that it cannot verify the developer: Blurry is not signed by Apple. Open System Settings › Privacy & Security, scroll down and click Open Anyway next to Blurry. You only do this once.

With Python (Windows, Linux, command line)

You need Python 3.12. The simplest way is pipx (or uv tool). With the desktop app:

pipx install "blurry-opsec[gui]"

Command line only:

pipx install blurry-opsec

The app

On a Mac (Blurry.dmg):

1. Drop photos and videos on the window, or use Choose Files… (⌘O).
2. A guide takes one file at a time: the file in the middle, and below it one choice at a time (sensitivity, cover, margin, and the sound for videos). From the cover step on, the picture shows exactly what will be exported; on a video you can play the covered result.
3. Correct the boxes by hand if needed: draw, move, resize or remove boxes on photos; in videos switch off a track that is not a face, or draw a still box over a span of time.
4. Export… opens the save panel on the original's folder, with <name>_blurry. The original is never touched. If no face was found, Blurry asks first.

Every file of the session is in View › Queue (⌘L). A short guide accompanies the first file (Help › Show the Guide Again).

On Windows and Linux, run blurry with no arguments (or blurry-app) to open the app installed with Python:

1. Drop photos and videos on the window, or use Choose files…. Each file is analysed in a separate process; nothing leaves your computer.
2. Review any file. Boxes found by the detector are yellow, boxes you add are teal. Drag on an empty area to add a box, drag a box to move it, drag its corners to resize it, press Delete to remove it. Preview result shows exactly what will be exported. In videos, scrub the timeline (uncertain moments are marked in red), turn off a track that is not a face, or draw a box that covers an area for a span of time.
3. Export. If a file has no face and you added none, Blurry asks before exporting it.

Both apps remember only a few settings (sensitivity, cover, margin, language, and on the Mac whether the guide was seen): never file names, folders or recent files. The Qt app uses its own file picker, because Qt's dialogs keep a list of recent folders; the Mac app uses the system's panels and removes what they record as soon as they close.

What it does

• Covers faces in photos and videos, using the YuNet face detector (bundled; its SHA-256 is checked on every start, and Blurry refuses to run if it does not match).
• Removes all metadata:
  • photos: EXIF (including the embedded thumbnail), GPS, XMP, IPTC, colour profiles and comments;
  • videos: container and track tags (QuickTime location, device, dates), chapters, subtitles, data tracks (such as GoPro and drone GPS) and cover art.
• Applies the rotation of phone photos and videos to the pixels first, so sideways faces are not missed, and the output looks the same as before without carrying rotation metadata.
• Drops audio by default, because voices can identify people. Keep it with --keep-audio.
• Never overwrites anything: the original stays untouched, and if the output name is taken a numbered name is used.
• Never stays silent when no face is found: it warns, and with --strict it writes nothing and exits with code 2.

What it does not do

Read the threat model before relying on Blurry. In short, it does not hide bodies, clothes, tattoos, voices, places, reflections, or the camera sensor's fingerprint; it cannot cover a face the detector does not find; it does not delete your original, which may also be synced to iCloud or Google Photos.

The detector can miss faces, especially faces in profile, in the dark, very small, or turned sideways. Always look at the result before sharing it. Files with uncertain detections are listed under flags in the --json report.

Uninstall

Mac: drag Blurry from Applications to the Bin. Blurry keeps no data: its only preferences (sensitivity, cover, margin, and whether the guide was seen) live in ~/Library/Preferences/com.chronocol.blurry.plist, which you can delete.

With Python: pipx uninstall blurry-opsec.

Report a problem

Something does not work, or a face is not covered? Open an issue: say what you did and what happened, without attaching photos or videos of real people. Security vulnerabilities do not go in public issues: follow SECURITY.md.

From the command line

blurry photo.jpg

writes photo.blurry.jpg next to the original.

blurry IMG_0001.HEIC clip.mov -o ~/Desktop/clean

writes IMG_0001.blurry.jpg and clip.blurry.mp4 in ~/Desktop/clean.

blurry INPUT... [-o FOLDER] [--level base|medium|high] [--mode solid|pixel]
    [--padding 0.25] [--keep-audio] [--no-faces] [--watermark TEXT]
    [--strict] [--json] [--debug]

Accepted files. Photos: JPEG, PNG, WebP (still), HEIC/HEIF. Videos: MP4, MOV, M4V, MKV, WebM, AVI. GIFs and animated images are refused. Photos come out as JPEG (PNG stays PNG); videos come out as MP4 (H.264).

Exit codes. 0 all good · 1 an error · 2 no face found with --strict.

Progress is printed on stderr as __PROGRESS__ {json} lines, for scripts and integrations.

Verify what you download

Every release on GitHub comes with a SHA256SUMS file, a CycloneDX SBOM, and a GitHub build-provenance attestation for each file, which proves it was built by this repository's release workflow.

• macOS / Linux, in the folder with the downloads:

  shasum -a 256 -c SHA256SUMS
  

• Windows (compare the result with the line in SHA256SUMS):

  CertUtil -hashfile <file> SHA256
  

• Provenance, with the GitHub CLI:

  gh attestation verify <file> --repo Chrono-Web/BLURRY
  

The PyPI package is published from the same workflow with Trusted Publishing (no tokens), and PyPI shows its provenance on the project page.

Known limits

• Faces in profile, in the dark, very small or turned sideways can be missed (see above).
• HDR videos from phones (10-bit HEVC) come out as 8-bit SDR H.264: colours can look flatter.
• On macOS, processing a video prints a harmless objc ... implemented in both warning, because OpenCV and PyAV each bundle their own FFmpeg.

Licence

Blurry is MIT-licensed. The PyPI package contains only Blurry's code, the YuNet model (MIT) and the Geist font (OFL). The Mac app (Blurry.dmg) also contains FFmpeg with x264 and x265, which are GPL-2.0-or-later, and the PyInstaller bootloader (GPL-2.0 with an exception for the programs it runs): see THIRD_PARTY_LICENSES.md for every licence and the exact sources.