Real, manipulated, or AI-generated?
Upload an image. Raven checks whether a generative model made it and, separately, whether a face was swapped in. It names the closest matching generator and scores the rest.
Beta preview. Available to verified agencies.
- 01Start with one frame, made with Higgsfield. At a glance, it reads as a real photograph.
- 02Pass one asks whether a generative model made the frame.
- 03The closest match is Other, at 92%. Every named generator scores near zero.
- 04Pass two asks, separately, whether a face was swapped into a real photograph. The signal is 2%.
- 05Verdict: likely AI-generated. Together, the two passes point to a fully synthetic frame rather than a doctored photograph.
Illustration. The two views are computed from the frame, not by Raven’s detector. Scores and verdict from the example below.
The result.
The frame was made with Higgsfield. Verify Image flagged it as likely AI-generated. The closest match was Other, at 92%, with every named generator near zero. The face-swap signal was 2%.
Read together, a 92% concentration on one generator and a 2% face-swap reading describe a fully synthetic frame rather than a doctored photograph.
What you get back.
A plain-language verdict with the working shown. A result resting on a narrow margin reads differently from one where the signal is overwhelming, and the scores make that visible.
- Verdict
- A plain-language judgment, such as likely AI-generated.
- Closest match
- The generator the signals most resemble.
- Per-tool confidence
- A score for each generator, so you can see how concentrated the result is.
- Face manipulation
- A separate reading for deepfake and face-swap edits.
Beta preview. Accuracy is not guaranteed. Verify Image is intended to flag frames that warrant closer scrutiny, not to serve as a standalone determination of authenticity. Treat results as investigative signal and corroborate through normal case procedure.
Check the frame before you trust the lead.
A synthetic image still produces a plausible location estimate. Find Region and Find Street will analyze a street that never existed. Running Verify Image first keeps an investigation from committing hours, and possibly an operational response, to a frame that was generated rather than captured.
Powered by Graylark's proprietary computer-vision models.
Common questions.
What does Verify Image detect?
Two separate things. Whether an image was generated by an AI model rather than captured by a camera, and whether a real photograph has been altered through a deepfake or face-swap edit.
Can it tell which AI tool made the image?
It gives a ranked view. When an image is judged synthetic, Raven names the most likely generator and scores the remaining candidates by per-tool confidence, so the result reads as a distribution rather than a single label.
Is Verify Image production ready?
It is a beta preview and accuracy is not guaranteed. It is intended to flag frames that warrant closer scrutiny, not to serve as a standalone determination of authenticity.
Why check authenticity before geolocation?
A synthetic image can still produce a plausible-looking location estimate. Verifying the frame first prevents an investigation from spending time on a scene that was never photographed.
To see where Verify Image fits in a case, read Geolocating a Suspected Stolen Car From a Single Social Media Post.
See Verify Image in Raven.
Book a walkthrough with your team. Raven is available to verified agencies and investigative teams.
