AI image metadata viewer

What hides inside an image file

A JPEG, PNG, WebP or HEIC file is more than a grid of pixels. Cameras and software write structured metadata into it, and several of these containers matter for AI detection:

EXIF

Written by cameras and phones: make, model, lens, exposure time, aperture, ISO, focal length, capture date and often GPS. Editing software adds a Software field. A rich, consistent EXIF block is typical of a genuine photo; generators almost never write one.

XMP

An extensible XML packet used by Adobe and many others. It can hold the creator tool, edit history, rights, keywords and, importantly, the IPTC extension field DigitalSourceType.

IPTC

The news-industry standard for captions, credits and keywords, now including the digital source type vocabulary that distinguishes camera captures from AI generations.

PNG text chunks

Key-value text stored in PNG files. Stable Diffusion tools write their prompt and settings here; ComfyUI stores the entire workflow graph as JSON.

C2PA manifests

Signed provenance records stored in their own boxes inside the file. See the C2PA viewer for every detail.

How GPTTrace reads it

When you drop an image, GPTTrace reads EXIF, XMP and IPTC with the exifr library, parses every PNG text chunk (decompressing zTXt and compressed iTXt chunks), extracts raw XMP packets so that fields exifr doesn’t know about aren’t lost, reads JPEG comment segments and walks any C2PA manifest. It then sorts what it found into fields that name the tool that made the file — Software, CreatorTool, generation parameters — and free-text fields such as descriptions and keywords, because a generator name in a caption is much weaker evidence than one in the Software field.

The result lists the evidence first, then a metadata snapshot of the most informative fields, then every PNG text chunk in full. A Stable Diffusion PNG will show its whole prompt; a phone photo will show its camera and exposure settings.

When there is no metadata

Most images you find online have been stripped by the platform that served them. That is normal and says almost nothing about whether a picture is real. In that case GPTTrace falls back to analysing the pixels with its neural classifier and forensic checks, and the metadata section simply stays short. If you need the metadata, try to obtain the original file from its creator or source rather than a downloaded copy.

Camera metadata that supports authenticity

A real phone photo usually has a dozen or more camera fields that agree with each other: a make and model, a lens that exists for that model, an exposure time and ISO that fit the lighting in the picture, a capture time and often GPS coordinates. GPTTrace counts these and reports “camera metadata present” when enough of them are there. A fabricated image with copied EXIF can fake this, but the fields often disagree — a lens from another brand, a night scene shot at 1/4000 s, a capture date before the camera model was released. If the details matter, check them against each other.