EXIF Data Explained: What Your Camera Writes Into Every Photo
Table of Contents
What EXIF Actually Is
Every photograph your phone or camera produces is more than pixels. Alongside the image data, the file carries a structured block of metadata called EXIF — the Exchangeable Image File format's information standard, introduced in the late 1990s for digital cameras. When you press the shutter, the device writes a small record describing the circumstances of the shot: what took it, when, with which lens settings, and often where.
EXIF lives inside the JPEG file itself, in a segment that begins right after the file header. For PNG files the
equivalent information is stored in tEXt chunks. Because it is plain structured data with no
cryptographic protection, anything that can write a file can write EXIF — cameras, phones, editing software,
scripts, or a tool like ours.
That last point is the one most people miss. EXIF is not a secure certificate of authenticity. It is a self-reported travel diary that the file carries around. Almost every provenance decision a platform makes — including the AI labels people try to get rid of — starts with reading this diary, which is why understanding it pays off.
EXIF describes how the file claims it was made, not how it was actually made. Stronger provenance systems (C2PA) exist precisely because plain EXIF proves nothing on its own.
The Fields That Matter
A typical camera JPEG carries between twenty and fifty EXIF fields. Most are technical noise, but a handful carry nearly all the signal:
- Make and Model (
Make,Model) — the manufacturer and device name, e.g.Apple/iPhone 15 Pro. The single most-read pair of fields. - Lens and focal length (
LensModel,FocalLength) — which lens was attached and at what zoom. A dedicated camera records the actual lens; a phone records its fixed equivalent (e.g. 24mm). - Exposure triangle (
ISO,FNumber,ExposureTime) — sensor sensitivity, aperture, and shutter speed. These three must be physically plausible together: an ISO 50 shot at 1/8000s in a dim room did not happen. - Timestamps (
DateTimeOriginal) — when the shutter fired, to the second. - GPS coordinates (
GPSLatitude/GPSLongitude) — present only if location permission was enabled. This is the field privacy tools strip first. - Software (
Software) — the last program that wrote the file. Cameras write firmware names; generators write their own names — which is exactly how files self-identify as AI. - Orientation and dimensions (
Orientation,PixelXDimension) — how the sensor was held and how large the output is.
What a Camera Writes vs. What a Generator Writes
Compare two files and the difference is immediate. A photograph taken on an iPhone has a dense, internally consistent EXIF block: Apple as make, a specific iPhone model, a main or telephoto lens at 24 or 77mm, a plausible exposure triple, a timestamp aligned with the file's creation time, and Apple's own software string.
A raw export from an image generator looks nothing like that. Depending on the tool it will have one of three profiles:
- An empty block. No camera fields at all. Suspicious in the opposite direction — real photos are never this clean.
- A self-incriminating block.
Software: Stable Diffusion, a prompt string, a sampler name, a model hash. Local Stable Diffusion setups famously write the entire generation recipe into the PNG. - A cryptographic manifest. Tools integrated with C2PA (DALL·E, Firefly, newer Midjourney exports) attach a signed manifest asserting the image was AI-generated. This is stronger than EXIF and harder to quietly remove.
The full provenance picture — and where SynthID fits into it — is covered in our guide to removing AI metadata; here we stay with classic EXIF because it remains the layer most tools read first.
Why Coherence Matters More Than Presence
The instinctive fix — delete everything — backfires, and the reason is worth internalizing. Detectors and platform classifiers do not merely ask is there metadata? They ask does this file's story hold together?
A real photograph's EXIF is coherent by construction because a physical device produced it in one act. The focal length matches the lens. The ISO matches the lighting the exposure implies. The timestamp matches the file's creation moment. The pixel dimensions match the sensor's output size.
An edited or synthetic file breaks that coherence in identifiable ways: a Canon model with an iPhone focal length, a timestamp hours before the file was created, a 3072×3072 square "photo" from a device that shoots 4:3 or 16:9, or — the classic — an empty metadata block on a file whose pixels show no sensor noise.
If you are going to manage a file's EXIF, the profile you write has to agree with itself and with the pixels. One contradictory field undoes the whole story. This is the principle the DEAI tool is built around: it never strips to empty — it replaces with a matched, internally consistent camera profile.
How to Read a File's EXIF Yourself
You don't need special skills to inspect metadata. In rough order of convenience:
- macOS Finder / Preview — right-click a file, Get Info, expand More Info, or
open in Preview and press
Cmd+Ifor the EXIF inspector. - Windows — right-click, Properties, Details tab. Shows camera, dimensions, and often GPS.
- ExifTool — the professional standard. Run
exiftool image.jpgfor a complete dump including C2PA manifest presence.exiftool -a -u -g1 image.jpgreveals even obscure groups. - Online viewers — several sites dump EXIF from an uploaded file. Convenient, but remember you are uploading your file (and its GPS data) to a stranger's server.
What to look for when assessing a file: an empty block (edited or generated), a Software field naming
a generator or an AI-capable editor, timestamps that precede file creation, dimensions that don't match any
claimed device, and C2PA/ JUMBF markers at the start of the file.
What Platforms Do With EXIF on Upload
Instagram, X, TikTok and most other major platforms do not keep your EXIF. During upload they read it once — extracting creation time and, if present, provenance claims — then re-encode the image and discard the metadata block entirely. They do this for privacy (GPS coordinates are a liability) and for file-size control.
This creates a one-way trip that matters strategically: the metadata read happens exactly once, at upload. A label applied from that read persists as platform-side state attached to the post — which is why removing metadata after posting does nothing, and why the only effective moment to manage a file's story is before it leaves your device.
Everything above describes how the system works. Whether you should adjust a given file's EXIF depends on context: correcting a false positive on your own photography is not the same act as disguising a fully synthetic image as photojournalism. We cover the ethics in the humanization guide.
Frequently Asked Questions
Do screenshots contain EXIF data?
No. Screenshots are rendered by the operating system, not captured through a camera sensor, so they contain no camera EXIF. This is one reason platforms treat screenshots differently from photographs when evaluating provenance.
Does Instagram or Twitter keep the EXIF from my uploads?
Generally no. Major platforms strip most EXIF during upload and re-encode the image — primarily for privacy (removing GPS) and file-size reasons. They read the metadata first, then discard it. That reading step is where AI labels usually originate.
Can EXIF data be faked?
Yes. EXIF is plain structured data with no cryptographic protection by default, so any tool can write arbitrary values into it. Only C2PA content credentials add signatures that make tampering evident. This is why serious provenance systems never rely on EXIF alone.
Why does my photo have no GPS data?
Most likely because location access for the camera app was off, or the platform you received it from already stripped it. Absence of GPS proves nothing either way — it is one of the most commonly removed fields.
EXIF is the foundation layer of image provenance — but it sits underneath two more sophisticated systems. To see how the full stack fits together, continue with how AI image detection works.
Process an image of your own
Strip AI provenance signals and inject a coherent camera profile in seconds — free, no signup.
Open the Tool