Was this image made by a camera or an AI?
Drop in any photo and SynthCheck reads its Content Credentials, IPTC origin tags, EXIF and generator metadata — then tells you exactly what the evidence does and does not prove.
- No upload — parsing is local
- No account, no limits
- Works offline
How the check works
Modern image files carry provenance data in several competing standards. SynthCheck parses the container itself — not a guess from the pixels — and reports every marker it finds.
1C2PA Content Credentials
The cross-industry provenance standard backed by Adobe, Microsoft, OpenAI, Google and camera makers. Stored as a signed JUMBF manifest recording the claim generator and every edit action.
2IPTC Digital Source Type
A one-field declaration in XMP. The value trainedAlgorithmicMedia means fully AI-generated; compositeWithTrainedAlgorithmicMedia means AI elements over other material. Used by Midjourney and Meta.
3Generator fingerprints
Stable Diffusion, ComfyUI, InvokeAI and Fooocus write the whole prompt, sampler, seed and model name into PNG text chunks. Nothing else on earth produces those fields.
4Camera EXIF
Body, lens, shutter, aperture, ISO, focal length and GPS. When enough of these agree, the file behaves like a genuine capture — useful supporting evidence, though EXIF can be forged.
What this can and cannot prove
Most "AI detectors" show you a confidence percentage invented by a classifier that was never calibrated. We would rather tell you the truth about the evidence.
Reliable
- Confirming AI origin when the generator labelled its own output
- Reading signed C2PA manifests and the actions inside them
- Surfacing prompts, seeds and model names left in the file
- Showing camera capture fields and GPS
- Catching a file whose extension lies about its contents
Out of reach
- SynthID. Google DeepMind's watermark lives in the pixels, not the metadata, and has no public reader. Use Google's own SynthID Detector.
- Stripped files. Screenshots and social re-uploads carry nothing to read.
- Verdicts from pixels alone. We do not run a guessing classifier.
- Certificate validation. We show the manifest; the official verifier checks the signing chain.
The rule that matters most
A clean file is not a real photo. Metadata is trivial to remove — saving a screenshot does it by accident. Absence of an AI marker is never evidence that an image is authentic. It only means this file has nothing left to tell you.
Markers are strong evidence in the positive direction and no evidence at all in the negative direction. Any tool that claims otherwise is selling you certainty it does not have.
Why this suddenly matters
Article 50 of the EU AI Act took effect on 2 August 2026. Providers of generative AI systems must mark synthetic output in a machine-readable format so it can be detected as AI-generated, with penalties reaching EUR 15 million or 3% of worldwide annual turnover.
In practice that means the markers this tool reads are moving from a voluntary nicety to a legal requirement, and the share of AI images that declare themselves is going up fast. Systems already on the market before that date have until 2 December 2026 to comply with the machine-readable marking obligation, and content generated before 2 August 2026 does not need labelling retroactively.
Deliberately stripping those markers to pass synthetic media off as real runs directly against the transparency regime — and against the terms of service of every major model provider.
Questions
Is my image uploaded to a server?
No. Every byte is parsed by JavaScript inside your own browser tab. There is no backend, no storage and no logging of file contents. You can disconnect from the internet and the tool still works — that is the easiest way to verify the claim yourself.
Can you detect SynthID from Google Gemini?
No, and be sceptical of anyone who says they can. SynthID embeds a statistical watermark in the pixels themselves, and the detector is not publicly available. SynthCheck only reads metadata. If a Gemini image still carries its C2PA manifest or IPTC tag we will find it, but if those were stripped the SynthID watermark may still be present and invisible to us. Google operates its own SynthID Detector for that.
Why do most images come back inconclusive?
Because Instagram, Facebook, WhatsApp, Telegram, Discord and X all re-encode uploads and discard most metadata in the process. Screenshots lose everything too. This is the normal outcome for anything saved off social media, and it is why a metadata-based answer has to be honest about its blind spots.
Can the markers be faked?
In both directions. EXIF can be copied from a real camera file onto a generated one, which is why camera evidence is reported as supporting rather than conclusive. C2PA manifests are cryptographically signed, so forging one convincingly is much harder — but validating the signing certificate requires the official verifier, which we link to on every result.
Which formats are supported?
JPEG, PNG, WebP, AVIF, HEIC/HEIF and TIFF. JPEG and PNG get the deepest parsing, including compressed zTXt and iTXt chunks. AVIF and HEIC get container-level box parsing plus a raw marker sweep.
What does it cost?
Nothing, and there is no account. The site is supported by advertising. Since no file ever leaves your device, there is no processing cost per check and no reason to meter you.