For journalists working on AI-generated media
If you are on deadline with an image or video you need checked, send it to us. We will run it and send back the per-signal evidence, including the signals that could not run and why. No cost, no account, no embargo.
Free forensic analysis, on deadline
Email the file or a link to support@fauxlens.com with "Press" in the subject line and your deadline in the first sentence.
Turnaround: within 24 hours as standard. If your deadline is sooner than that, say so in the subject line and we will tell you quickly whether we can meet it rather than leaving you waiting.
We will send you the verdict, the confidence score, which forensic signals fired, which ones could not run on that file and why, and a plain-English reading of what that combination does and does not support. If the honest answer is "inconclusive," that is what you will get.
Why we do not publish an accuracy percentage
Most detectors lead with one. We deleted ours, and we would rather explain that than have you find out later.
A single figure averaged across every generator, image quality and post-processing history hides exactly the variation that decides whether a result is usable in a given story. The published evidence is blunt about this: Deepfake-Eval-2024 measured image-detection AUC falling by roughly 45% on real-world media compared with academic benchmarks, and the winning entry in Meta's Deepfake Detection Challenge scored 82.56% on its public set against 65.18% on unseen generators.
So we show you the signals instead. Our methodology page documents how each one works and the conditions under which each becomes unreliable - re-compression, social-media re-encoding, heavy post-processing, and generators tuned specifically to evade detection.
If you are comparing tools and want a number from us for a table, the honest answer is that we will give you per-condition results on a named test set rather than one headline figure. Ask and we will send what we have, including where we lose.
Boilerplate
FauxLens is a free forensic tool that checks whether an image or video was generated or edited by AI.
FauxLens is a free forensic media authentication tool. It runs multiple independent forensic signals over an image or video - compression history, camera sensor noise, generator frequency artefacts, lighting physics, and embedded provenance metadata - and reports what each one found, rather than returning a single verdict. Uploads are deleted the moment analysis ends and are never used for training.
Founder
Netanel Ossi built and maintains FauxLens. He is a backend engineering manager by trade and writes the detection pipeline, the API and the site itself. He is available for interview on how forensic detection works, where it fails, and why confident-sounding accuracy numbers in this category deserve scrutiny.
Contact: support@fauxlens.com
How to cite a FauxLens result
Please attribute a specific scan rather than the tool in general, and include the date - the pipeline changes, and a result from March is not a result from today.
"A forensic analysis by FauxLens on [date] found [verdict] at [confidence] confidence, with [n] of its forensic signals indicating [finding]."
Please do not write "FauxLens confirmed the image is fake." The tool reports evidence and a confidence score; it does not certify authenticity, and the confidence score is a weight-of-evidence measure rather than a calibrated probability. A score of 92% does not mean 92 out of 100 such images are AI-generated.
- Tool
- FauxLens
- URL
- https://fauxlens.com
- Founder
- Netanel Ossi
- Methodology
- https://fauxlens.com/methodology
Logos, screenshots and anything else
Email support@fauxlens.com and say what you need - logo files, a screenshot of a specific report state, or a walkthrough of the pipeline for a technical piece. Same 24-hour turnaround.