Was this image AI-generated?
Flagged a suspected AI image? Screen it here to check for verifiable provenance and capture a documentable record. We read its Content Credentials (C2PA), file metadata, and known AI-tool markers, then give you one clear AI-generation likelihood score. We surface the signals — you make the editorial call. The image is analysed entirely in your browser and never leaves your device.
Drop an image to screen
or click to browse · JPEG, PNG, WebP, TIFF, HEIC, AVIF, GIF, SVG
For the strongest result, screen the author's original image files — exporting a figure into a manuscript usually strips the credentials and metadata this tool reads.
Technical details (for the record)
the raw signals behind the summary aboveAnalysing…
File metadata (EXIF)
unverifiedTechnical info cameras and apps embed (device, date, GPS, software). Easily edited or stripped — a lead, not proof.
All recognised tags
EXIF/XMP data is embedded by the originating device or app and can be edited by any tool. Use it as informational context — only Content Credentials (C2PA) are cryptographically verifiable.
Content Credentials (C2PA)
A tamper-evident, cryptographically signed record of how the image was made — the most reliable evidence here.
Recorded actions
| Action | Software | Digital source type | When |
|---|
Validation
SynthID check
unverifiedSynthID is Google DeepMind's invisible watermark for AI-generated content. The actual watermark can only be verified through Google's hosted SynthID Detector portal; this app only looks for textual mentions of "SynthID" inside the file's metadata.
How we map evidence to a verdict
- AI-generated — manifest declares a trained-algorithmic digital source type.
- AI involved — composite, AI-enhanced, or signed by a known generative-AI tool.
- Human-captured — manifest says the image was captured by a camera (or only edited by a human) with no AI markers in the chain.
- No C2PA — image carries no Content Credentials. Absence is not proof either way.
- Unknown — manifest exists but lacks the standard provenance fields we need.
A valid Content Credential is signed, verifiable evidence. Metadata hints are leads, not proof. Absent credentials ≠ a genuine, human-made image.