Privacy

EXIF metadata: what your photos reveal

A photo is not just an image: it is a document with a dossier stapled to the back. Inside the file, a metadata block called EXIF records which camera took it, with which lens, at what instant, at what aperture, and, if GPS was on, the exact point on the planet. None of this shows on screen; all of it travels when you send the original file. This guide opens that dossier field by field, shows what social networks strip and what they keep, and explains why deleting EXIF does not make an image anonymous. At the end, you clean your photos in the [EXIF remover](tool:remover-exif) without them ever leaving your browser.

J-Kit12 min readBeginner
  • EXIF
  • Privacy
  • Metadata
  • GPS
  • Image

Key takeaways

  • EXIF (CIPA DC-008 standard, Exif 3.0) lives in a JPEG APP1 segment and stores camera, date, settings and, when present, GPS coordinates.
  • Most large platforms strip EXIF when they re-encode the image, but that is a side effect of compression, not a guarantee, and the original often stays on their servers.
  • Deleting EXIF does not anonymize: the content, the sensor noise (PRNU) that fingerprints the camera and sometimes an embedded thumbnail of the original all remain.

The photo is a document with a dossier

EXIF stands for Exchangeable image file format. It is the standard that describes how a camera records information about the capture inside the image file itself. Today it is maintained jointly by CIPA and JEITA under the CIPA DC-008 specification, the current version is Exif 3.0, published in 2023, which added UTF-8 text support. In a JPEG, this data lives in an application segment called APP1 (marker 0xFFE1), right at the start of the file; in a TIFF, it sits in image sub-directories (IFDs) pointed to by a private tag. Either way, the bytes travel glued to the pixels: whoever gets the original file gets the dossier.

The best way to understand is to read a real metadata dump. Below is a reading in the style of the exiftool utility, the values are made up, but the format is exactly what a common smartphone produces. Notice how many lines describe you, not the photo.

$ exiftool DSC_04821.jpg      # exemplo ilustrativo, valores ficticios

Make                : Apple
Camera Model Name   : iPhone 14 Pro
Lens Model          : iPhone 14 Pro back triple camera 6.86mm f/1.78
Software            : 17.5.1
Date/Time Original  : 2026:03:14 18:42:07        <-- quando / when
Exposure Time       : 1/120
F Number            : 1.8
ISO                 : 64
Orientation         : Rotate 90 CW              <-- como foi segurado
GPS Latitude        : 40 deg 44' 54.36" N       <-- onde / where
GPS Longitude       : 73 deg 59' 08.36" W
GPS Altitude        : 11.3 m Above Sea Level
GPS Date/Time       : 2026:03:14 22:42:05Z
Serial Number       : (alguns corpos gravam; MakerNote)
Thumbnail Image     : (Binary data 8734 bytes)  <-- miniatura embutida
Annotated EXIF metadata dump. Illustrative example with fictional values.

Three lines say when (Date/Time Original), where (GPS) and with what (Make, Model, Lens, Serial Number). Together they turn a cat photo into a stamp of time and place. And EXIF is not the only possible metadata container inside an image, just the most common one in cameras.

EXIF
Capture metadata written by the camera: date, settings, lens, orientation and GPS. The focus of this guide.
IPTC
A photojournalism standard: caption, credit, keywords, headline. Describes the content, not the machine.
XMP
Adobe’s XML wrapper (2001). Holds edits, notes and copies of EXIF/IPTC fields under namespaced names like dc:creator.

What exactly rides along

Not every field is sensitive: aperture (F Number) or exposure time interest photographers and no one else. The problem is the fields that point back at you. The table gathers the most revealing ones.

The EXIF fields that most expose the photo’s author.
FieldExampleWhat it reveals
GPS Latitude/Longitude40°44′54″N, 73°59′08″WThe exact point of capture, home, work, school.
Date/Time Original2026:03:14 18:42:07The exact instant, routine, presence, timeline.
Make + ModelApple · iPhone 14 ProThe device; links photos from different sources to the same owner.
Serial Number (MakerNote)(written by some bodies)Ties every photo to one specific physical camera body.
Software17.5.1 / Photoshop 26.0The OS version and whether, and how, the image was edited.
OrientationRotate 90 CW (value 6)How the device was held; why photos “flip” on their own.

The coordinate is not stored as a friendly decimal number. GPS lives in its own sub-directory, the GPS IFD, with dozens of standardized tags: latitude and longitude go in as three fractions, degrees, minutes and seconds, plus a reference tag (N/S, E/W) and another for altitude. Worked example: a latitude stored as [40/1, 44/1, 5436/100] means 40 degrees, 44 minutes and 54.36 seconds. Converting, 40 + 44/60 + 54.36/3600 = 40.7484° N, a precision of a few meters, enough to land on your front door on a map.

3 fractionsdegrees, minutes and seconds per coordinate
~meterstypical smartphone GPS precision
8 valuesthe Orientation tag (1 to 8): rotations and flips

The Orientation tag explains an everyday mystery. The sensor always records pixels in the same physical direction; when you turn the phone to portrait, it does not rotate the matrix, it just writes a number from 1 to 8 telling the viewer how to rotate or mirror the image. Value 6, for instance, means “rotate 90° clockwise”, the most common case for a vertical phone photo. If a program ignores that tag, the photo shows up sideways. It is harmless metadata, but it shows clearly that the image you see is already an interpretation of the bytes, not the raw bytes.

Where this habit came from

Embedding data inside the photo is old, but geolocation by default is recent, and it came from convenience, not surveillance. The timeline shows how we got here, and where the movement now points.

  1. 1995Exif 1.0 (JEIDA)

    The first version of the standard defines the image format and attribute tags, maker, model, date and capture settings.

  2. 1997–1998GPS enters the standard (Exif 2.x)

    The Exif 2.x line adds the sRGB color space, compressed thumbnails and the GPS IFD, the compartment that would later carry coordinates.

  3. 2001Adobe’s XMP

    Adobe introduces XMP, an XML wrapper for metadata that editing software uses to record authorship and history alongside EXIF.

  4. 2008 onwardThe GPS smartphone era

    GPS phones (starting with the iPhone 3G, 2008) make geotagging a default: billions of photos are now born with coordinates, most without the owner noticing.

  5. 2021–2024C2PA / Content Credentials

    A coalition (Adobe, Microsoft, BBC, Google, Sony and others) standardizes cryptographically signed provenance metadata, the opposite movement: instead of hiding the origin, proving the origin.

C2PA deserves a paragraph because it inverts this article’s logic. Where EXIF is passive metadata you want to remove, Content Credentials is active, signed metadata you may want to keep: it records origin and edit history in a manifest that, once signed, exposes any later tampering. It is the difference between a dossier that leaks your location and a seal that proves where the image came from, useful in the age of AI-generated images.

What platforms strip, and what they do not

The good news: most large social networks re-encode the image you upload, to compress and standardize it, and that re-encoding throws EXIF away as a side effect. Facebook, Instagram, X and others publish the clean version to the public. The bad news has three parts. First: that is a consequence of compression, not a privacy guarantee, and policies change without notice. Second: “stripping the public version” is not the same as “deleting”; several platforms keep the original file, with all its EXIF, on their own servers. Third: there are shortcuts that preserve everything, sending a photo as a “document” or “file” instead of an image usually skips compression and delivers the untouched original on the other end.

Keep in mind the photo is not the only leak in a post. The IP address of the connection that made the upload also identifies your network, the public vs. private IP guide explains what it reveals and what it does not. Image metadata is just one of the layers.

Why deleting EXIF is not anonymity

Here is the fine point almost every tutorial forgets: removing EXIF cleans the label, not the image. A photo with no metadata still carries, in the pixels themselves, enough information to locate and identify. Four residues survive the cleanup, each one demolishes the idea that “I stripped the EXIF, therefore I am anonymous”.

Sensor noise (PRNU) fingerprints the camera

Every sensor has tiny manufacturing imperfections that stamp a fixed, unique noise pattern onto every photo, the so-called PRNU (Photo-Response Non-Uniformity). The seminal work of Lukáš, Fridrich and Goljan (2006) showed this “noise fingerprint” can tie a photo to the specific body that took it. Worse: the pattern usually survives lossy compression and resizing, so re-encoding the image does not erase it.

The content gives itself away

License plates, storefronts, mountains in the background, a reflection in a window, the pattern of a curtain, the scene itself is often enough to geolocate. With no metadata at all, a street photo can still be cross-referenced against satellite or map imagery until it reveals where it was taken.

The embedded thumbnail may preserve the original

EXIF can carry a thumbnail of the photo. If an editor updates the large pixels but does not regenerate that thumbnail, the small version keeps showing the pre-edit image, including a face you blurred or a plate you censored. It is a classic leak, and the reason post-removal verification has to include the thumbnail.

The copies already published are still out there

Cleaning the file on your computer does not recall what already left. If you posted the original version earlier, someone may have downloaded it with the full EXIF, and caches, backups and third-party servers keep copies beyond your reach. Metadata, once published, is nearly impossible to revoke.

How to remove metadata for real

Safe removal has four steps, and the one most people skip is verification. Trusting a “remove metadata” button without checking the result is like closing the door without testing the handle.

  1. Verify firstOpen the photo in an EXIF reader and see what is there: GPS? date? serial number? embedded thumbnail? You cannot safely remove what you do not know is present.
  2. Remove by re-encodingRe-encoding the image (redrawing it into a new file) discards the whole EXIF block, including GPS and the thumbnail. It is more reliable than deleting tag by tag, because it leaves no residual container behind.
  3. Verify again, including the thumbnailReopen the clean file in the reader. Confirm GPS, date and serial are gone and that there is no embedded Thumbnail Image. This is the step that closes the thumbnail leak.
  4. Remember the copiesCleaning the new file does not erase versions you already shared. If the original leaked, treat it as public, removal protects the next post, not the previous one.

One last useful framing: metadata is not the same as the image, just as a hash is not the same as the data it summarizes. If that border between “the content” and “what describes the content” interests you, the hashing, cryptography and encoding guide works the same idea on different ground. In the end, image privacy is deciding, consciously, what stays inside the file before it leaves your hand.

Frequently asked questions

Do Instagram and Facebook remove EXIF from my photos?
In practice, most large networks re-encode the image when publishing it, and that discards EXIF from the version the public sees. But it is a side effect of compression, not a contractual guarantee, the policy can change, and the original file with all its metadata usually stays on the platform’s servers. To be safe, strip the metadata before uploading, on your own device.
Does removing EXIF make the photo anonymous?
No. EXIF is only the label. After removing it, you still have the scene content (plates, storefronts, reflections that geolocate), the PRNU sensor noise that can identify the specific camera and, sometimes, an embedded thumbnail of the original. Anonymizing an image is far more than deleting metadata.
Why do my photos appear sideways in some programs?
Because of the Orientation tag. The sensor always records in the same physical direction and just writes a number from 1 to 8 saying how to rotate or mirror the image on display, value 6, for instance, is “rotate 90° clockwise”. Programs that honor the tag show the photo correctly; those that ignore it display the raw, sideways pixels.
I blurred a face in the photo. Is that enough?
Not always. If the program does not regenerate the embedded EXIF thumbnail, the small version may keep showing the original, unblurred face. When covering sensitive information with a tool like the image censor, also verify that the old thumbnail was removed, covering the large pixels is not enough if the thumbnail preserves the original.
Is it safe to use a website to remove metadata?
It depends on where the photo is processed. A tool that uploads the image to a server to “clean” it is receiving the very file, with all the metadata you want to hide. Prefer 100% client-side tools that read and re-encode the photo in the browser itself, like this site’s EXIF remover, where the image never leaves your device.

Every photo ships with a dossier: EXIF records camera, date and, when GPS is on, the exact point of capture. Most networks remove it when they re-encode, but by accident, not by promise, and the original may stay on the server. Strip it yourself, before uploading, and verify the result (including the embedded thumbnail). And do not confuse cleaning metadata with anonymity: the content, the sensor noise and the already-published copies keep speaking for you.

Sources & references

  1. CIPA, Standards (DC-008, Exif)
  2. IPTC, Exif 3.0 released featuring UTF-8 support
  3. Exiv2, Standard Exif tags table (incl. GPS IFD)
  4. Lukáš, Fridrich & Goljan (2006), Digital camera identification from sensor pattern noise (IEEE TIFS)
  5. C2PA, Coalition for Content Provenance and Authenticity