Free Deepfake Video Detector

Deepfake or fully AI generated: what the words mean

People use "deepfake" for almost any fake video now, which muddies things. It helps to separate two cases.

A deepfake, in the older and narrower sense, starts with real footage and changes a real person in it: their face, their voice, what they appear to say. The room, the lighting and the camera are often genuine. Only the person has been tampered with.

A fully AI generated video has no camera footage behind it at all. A model produced every frame from a text prompt or a still image. The person in it may look like someone real, or may not exist.

The distinction matters because it changes where you look. With an altered person, the background can be perfectly normal and the problem is concentrated in one face. With a fully generated clip, the tells can be anywhere.

What this check reads

The checker reads the file for evidence it carries about itself: Content Credentials under the C2PA standard, an AI label that uses the IPTC digital source type codes, or the name of an AI video tool in its metadata. What it finds becomes one of four verdicts: AI generated, Likely AI, Likely real, or No AI signs found. It also pulls still frames from across the clip for you to look at. It does not score them.

Be clear about what that means for a suspected deepfake. Deepfakes made to deceive rarely carry a label that admits it, so "No AI signs found" is the result you should expect from most of them. The check was not built to spot a replaced face or a mouth re-synced to new audio, and it does not tell you who the person in the clip really is. The frames are where you do that work: step through them, and when one shows the face, look closely at it.

Signs worth checking by eye

Pause the clip on a frame where the face is large and well lit, then step through slowly. These are the spots where altered faces most often give themselves away.

• The edge of the face. Look along the jaw, hairline and ears. Blurring, a faint seam, or skin that does not match the neck can point to a face that was laid over another.
• The mouth while talking. Teeth that smear into one block, lips that do not quite close on "m" and "b" sounds, or a mouth that looks sharper or softer than the rest of the face.
• Eyes and glasses. Reflections that differ between the two eyes, or glasses frames that bend or vanish as the head turns.
• Lighting on the face versus the room. A face lit evenly from the front in a room with one hard side light is worth a second look.
• The voice on its own. Close your eyes and listen. Flat emotion, odd breathing, or a voice that does not fit the room can matter as much as the picture.

For more tells with examples, see how to spot AI video and the step by step method in how to tell if a video is AI. To check a voice note or a photo of the same person as well, DeepfakeDetector.ai handles images and voice, and its free account covers 50 scans a month on clips up to 2 minutes.

What to do if a clip of you is fake

Finding a fake video of yourself is unsettling. Some practical steps:

1. Save evidence first. Keep the link, a copy of the file if you can get one, the account name that posted it, and the date you found it. Posts can disappear once reported.
2. Report it on the platform where it lives. On YouTube, the Protecting your identity page explains that you can ask for removal of AI generated or other synthetic content that looks or sounds like you. YouTube says it weighs factors including whether the content is realistic, whether it is disclosed as altered or synthetic, whether you can be uniquely identified, and whether it has parody, satire or public interest value. The complaint has to come from you or your legal representative.
3. Tell the people it might fool. If the clip asks for money or makes claims in your name, a short message to family, colleagues or followers can stop the damage faster than any takedown.
4. Get real help if it is threatening or sexual. Contact local police or a legal adviser, and keep your saved evidence ready.