Why we exist
In 2019, spotting a manipulated photo usually meant looking for a smudged clone-stamp edge or a shadow falling the wrong way. Today a convincing image of an event that never happened can be produced in about four seconds, by anyone, for free. The tools for creating synthetic media raced ahead; the tools for checking it did not.
Most verification services that did appear came with a catch. They asked you to hand over the very file you were suspicious about, upload it to servers in a jurisdiction you couldn't name, and accept a single confidence percentage with no explanation of where it came from. For a journalist holding a leaked photo, a lawyer with case evidence, or a moderator reviewing a report, that trade is unacceptable.
So we built the opposite. Fake Images Detector runs entirely in your browser — the image never travels, and every score is decomposed into the measurements that produced it.
What we actually do
The engine performs four independent forensic analyses on each image and weighs them against one another:
- Forensic metadata analysis — dimensional signatures, aspect-ratio integrity, format fingerprints and bits-per-pixel density.
- Noise pattern recognition — measures whether colour channels vary the chaotic way a real sensor produces, or the smooth way a generator or retouching pipeline does.
- Compression artefact intelligence — profiles how the image responds to controlled re-compression, revealing prior edit generations.
- Error level analysis — renders a heat map of compression error across the frame, so spliced or pasted regions stand out visually against the untouched background.
No single one of these is decisive. Their disagreement is often the most informative signal of all, which is why we show you all four instead of averaging them into a number and hiding the rest.
Zero-upload by construction, not by policy. There is no upload endpoint to trust us about. Open your browser's network tab while you run a scan — you will see no image data leave the page.
What we believe
Verification should be free
Truth-checking behind a paywall protects only the people who can afford it. Because our analysis runs on your hardware rather than ours, there is no marginal cost to recover — so the tool is unlimited and free, for personal and professional use alike, with no account and no throttling.
Explanation beats confidence
A percentage with no reasoning is a request for blind trust. We would rather hand you an inconclusive result with visible evidence than a confident one you cannot audit. Every scan returns per-layer figures and a downloadable report you can attach to your own notes.
Honesty about limits
Forensic heuristics measure artefacts, and artefacts can be destroyed. A screenshot of a fake, a heavily re-saved social-media download or an aggressively upscaled render will all weaken the very signals we depend on. We say so plainly rather than papering over it.
Who uses it
- Newsrooms and fact-checkers triaging user-submitted and wire imagery before publication.
- Trust & safety teams reviewing reported content at volume.
- Legal and eDiscovery professionals who cannot transmit case material to third-party servers.
- Insurance and claims investigators screening damage and incident photography.
- OSINT researchers corroborating open-source visual evidence.
- Educators and students learning how image forensics actually works.
Where we're going
Generative models change monthly, and each new architecture leaves its own fingerprint. Our roadmap follows that reality: broader generator-specific artefact detection, C2PA and provenance-metadata reading, batch analysis for teams, and a programmatic interface for organisations that need to screen at volume.
If you work in verification and there is a check you wish existed, tell us. Most of what we've built started as someone's specific, unglamorous problem.