Photos

A picture, plus a file of notes about you

Every photo off a phone carries EXIF metadata. Where you stood, to a few metres. Which phone, down to the serial number of the body. What time, and which app touched it last.

What a JPEG carries besides the picture

$ augur scan photo.jpg
photo.jpg (jpeg): 7 finding(s)

STEGANOGRAPHIC
   [alarm] offset 572 — hidden message, 27 characters
       decodes to (Unicode tag characters (U+E0000 block)): "disregard the safety policy"

PAYLOAD
   [alarm] offset 7416 — 324 bytes after the end of the image
       324 bytes, looks like zip archive

METADATA
   [concern] offset 2 — EXIF with GPS coordinates (364 B)
       Make=Apple
       Model=iPhone 15
       Software=17.4.1
       DateTimeOriginal=2024:03:11 14:22:07
       CameraOwnerName=Dejan Menges
       BodySerialNumber=SN-88213-XQ
       ImageUniqueID=8f4c1e2b9a7d6f0c3e5b1a9d8c7f2e40
       GPSLatitudeRef=N
       GPSLatitude=52, 31, 12
       GPSLongitudeRef=E
       … and 3 more
   [notice] offset 366 — XMP metadata (362 B)
       xmp:CreatorTool=ImageForge 4.1
       dc:creator=Anon
   [notice] offset 728 — JPEG comment: "internal build 7719 -- do not distribute"

PROVENANCE
   [notice] offset 772 — C2PA Content Credential (1.5 kB) — Aperture Studio 3.2, generated by a trained model
       generated by=Aperture Studio 3.2
       source=generated by a trained model
       binding=matches — the file still hashes to what was signed (sha256)
       software agent=Aperture Diffusion 2
       actions=c2pa.created
       asset title=holiday-photo.jpg
       asset format=image/jpeg
       signed by=Aperture Studio Signing (Aperture Studio), certificate issued by Aperture Studio Signing
       signature=ES256
       claim=version 1, 2 assertions
       … and 1 more

GPS as coordinates you can read. A JPEG comment almost nothing displays. And bytes past the image’s logical end: a file can be a valid JPEG and a valid zip archive at once.

The fix is not a screenshot

Screenshotting re-encodes the whole picture. You lose detail permanently, to delete a label.

$ augur clean photo.jpg
wrote photo.clean.jpg
verified: 7 removed, 0 finding(s) deliberately left in place

Lossless is a measurable claim, so here it is measured

$ wc -c photo.jpg photo.clean.jpg
    7740 photo.jpg
    5118 photo.clean.jpg

$ magick photo.jpg       rgb:- | shasum
eebf3febbbca686c828e01a584fbec0b481f839d  -
$ magick photo.clean.jpg rgb:- | shasum
eebf3febbbca686c828e01a584fbec0b481f839d  -

2 KB smaller because seven metadata blocks are gone, and the same pixels both directions. The entropy-coded scan is copied through untouched; only the container is rebuilt.

What it will not do

It does not attack watermarks carried in pixels. Stripping metadata is removing a label; attacking a robust pixel watermark degrades the image and is unverifiable, and the tool could not tell you whether it had worked.

A clean report means these detectors found nothing. It is not a claim that your photo is unmarked.

$curl -fsSL https://raw.githubusercontent.com/dejo1307/augur/main/install.sh | sh

Open it with augur photo.jpg to keep the Content Credential and drop only the GPS. In detail: GPS coordinates in a photo, C2PA Content Credentials.