Deepfake Detector: Is This Video, Photo or Voice Real?

Which deepfake detector do you need: image, video or voice?

Pick the detector that matches what you received. One account covers all three, and the same credits work on each.

When you are unsure which part matters, start with the part that carries the request. In a video message asking for money, the voice can matter as much as the face, and the video check covers both.

How does deepfake detection work across images, video and voice?

Each medium goes through the same four-step method, with a detector specialised for that medium.

1. 01Provenance first: C2PA content credentials and EXIF/XMP metadata are read. If they name a known AI generator, the verdict is probably AI and the reason is stated.
2. 02An open-source detector scores the content: an AI-image classifier for photos; frame scoring plus a face-swap classifier for video; an anti-spoofing model for voice.
3. 03In the grey zone, Claude reviews the content as a second opinion. It can move the score only a little and never decides alone.
4. 04Claude writes the explanation in English, French or Arabic, with the signals and the limits of the method.

A video combines the other two: its frames go through the photo detector and its soundtrack through the voice detector. That is why a video can come out with real images and a synthetic voice.

A deep fake check you can run before using any tool

Most deepfakes are used to make you believe or do something. These questions catch many of them before any detector runs.

1. 01Source: who posted it first, and is that account who it claims to be?
2. 02Context: do the date, place and circumstances add up?
3. 03Request: does it push you to pay, share credentials, hurry or keep a secret?
4. 04Channel: can you confirm with the person through a channel you already trust?
5. 05Details: zoom on faces, hands and text; listen for a voice that does not match the lips.
6. 06Corroboration: has anyone reliable reported the same thing independently?

If any answer worries you, stop and verify before you act, whatever a detector later says. A clean result does not cancel a suspicious request.

What a deepfake detection tool cannot do

A detector estimates whether content was generated or manipulated. Several important questions stay outside its reach.

• It does not identify who made the fake or who is being imitated.
• It cannot confirm that a genuine video is shown in its true context. Real footage with a false caption will come out as probably human.
• It cannot judge whether a real person is telling the truth.
• Small manipulations inside otherwise real media, and heavily degraded files, often land in the uncertain zone.
• It does not watch live calls or streams; it works on files and links you submit.

Identity checks are where this matters most. Our page for fraud and KYC teams explains how to use detection as one layer among others.