Free AI Image Metadata Viewer
Check what is hidden inside an image: C2PA Content Credentials, the IPTC AI-generation label, generator signatures, GPS coordinates, and Stable Diffusion prompt data. Nothing is uploaded.
What a result looks like
A real image from Gemini, and the unedited output this tool gives for it.
Gemini_Generated_Image.png
1.2 MB · 1024 × 1024 · PNG
AI metadata detected
Metadata can be edited or removed, so treat this as a signal rather than proof.
Evidence
- Digital source type
trainedAlgorithmicMedia
Generated entirely by a trained AI algorithm. Read from the C2PA claim.
- Signed by
Google C2PA Core Generator Library
- Recorded actions
c2pa.created, c2pa.edited, c2pa.opened, c2pa.converted
- Content Credentials
2 manifests, signed, 1 ingredient
Derived from another credentialed asset. Signature validity is not checked.
Below this, the full report continues with the raw JUMBF box tree and the file's technical tags: 18 fields in total for this image.
What these tools detect
Four categories of embedded data, all read locally in your browser.
- C2PA Content Credentials
- Cryptographically signed provenance records embedded by OpenAI, Google, and Adobe Firefly. We report whether a manifest container is present; we do not verify its signature.
- IPTC DigitalSourceType
- The cross-industry machine-readable AI label, for example trainedAlgorithmicMedia. Adopted by OpenAI, Google, Meta, Microsoft, Adobe, and Shutterstock.
- Generation parameters
- Prompts, seeds, samplers, CFG scale, and model hashes that Automatic1111 and ComfyUI write into PNG text chunks, and that local, on-device apps such as Draw Things write into ordinary EXIF/XMP fields like Description, as plain, readable text.
- Generator signatures
- Tool names left behind in ordinary tags such as Software or Credit, covering Midjourney, GPT Image 2, Nano Banana, Firefly, Imagen, Runway, Leonardo, Ideogram, FLUX, and local tools like Draw Things and Fooocus.
Metadata is not a watermark
These tools read the data wrapped around an image, meaning EXIF, XMP, IPTC, and C2PA. They never analyze the pixels. Invisible watermarks such as Google's SynthID are encoded into the pixel data itself, so they survive metadata stripping, format conversion, resizing, and screenshotting.
No third-party tool can read SynthID either. Detection is available only through Google's own Gemini app and Cloud partners, so any site claiming to detect it from metadata alone is overstating what it does.
A visible watermark is pixel data too, which is why stripping metadata never removes one. Those have to be corrected in the picture itself: our Gemini Watermark Remover does that for the sparkle logo Gemini adds, by solving the blend that drew it rather than painting over it.
The practical consequence: an image showing no AI metadata here may still be AI-generated, and an image cleaned with our remover may still carry an invisible watermark.
Fix the yellow tint while you are here
- One-click yellow tint removal
- Fix up to 10 images at once
- Results in under 2 seconds
- 3 free credits on signup
Frequently asked questions
- Does this tool tell me for certain if an image is AI-generated?
- Not with certainty. It surfaces signals that many AI tools embed, like the IPTC DigitalSourceType label or a C2PA Content Credentials manifest, but plenty of AI generators (Midjourney, most Stable Diffusion frontends) don't embed those, and any metadata can be stripped or edited. Absence of these signals doesn't mean an image isn't AI-generated.
- Why does my image show no metadata at all?
- Most social platforms strip metadata on upload to save space and protect user privacy. A file that has passed through Instagram, X, or WhatsApp will usually arrive with nothing embedded, regardless of how it was originally created.