A solo iPhone deepfake detector shows smart product choices
An iOS app now checks whether a photo or video is real, made by AI, or a deepfake. It can analyze media from the phone’s library or camera, or inspect a pasted social link, then return a trust score with the reasons behind it. The main privacy choice is that the check runs on-device.
Nothing is sent to a server, there is no cloud processing, no account, and no tracking. The developer trained a CoreML model to tell real photos apart from AI-made images. The hardest problem was avoiding , especially real photos being marked as fake, and the reached about 1.6% on a separate test set.
The app does not give only a simple yes or no. It checks several signals, including , metadata, AI-detection clues, frequency analysis, and time in video, then combines them into one score. When the evidence is weak, it leaves the result inconclusive instead of making a risky claim.
Key points
- The app checks photos, videos, and pasted social links, then gives a trust score.
- All analysis runs on-device, with no upload, cloud account, or tracking.
- A self-trained CoreML model separates real photos from AI-made images.
- The reported is about 1.6% on a separate test set.
- Weak evidence leads to an inconclusive result instead of a forced answer.