Deepfakes Explained: How Synthetic Media Is Made and Spotted

The term deepfake combines “deep learning” and “fake”. It describes images, audio or video altered or generated by AI to show someone doing or saying something they did not. Some uses are harmless or creative; others deceive or harm.

Common types

• Face swap: one person’s face is mapped onto another’s body in a video.
• Lip-sync manipulation: mouth movements are changed to match new audio.
• Voice cloning: synthetic speech imitating a particular person.
• Fully generated images: realistic people or scenes that never existed.

How they are made, in outline

Systems learn the look or sound of a subject from sample material, then generate new content that matches. Techniques include autoencoders, adversarial networks and diffusion models. Tools have become more accessible, so producing a rough fake no longer requires specialist skills, though polished results still take effort.

Legitimate uses

• Film production, such as de-aging or dubbing actors into other languages.
• Accessibility, such as restoring a voice for someone who lost it, with consent.
• Education and museum exhibits with clear labeling.
• Satire and art, when presented honestly.

Harms

• Non-consensual intimate imagery, a serious form of abuse.
• Fraud, such as impersonating an executive’s voice to request a payment.
• Political disinformation and reputation attacks.
• The “liar’s dividend”: real evidence dismissed as fake.

Clues that something may be synthetic

No single sign is reliable, and quality improves over time. Still, these cues sometimes appear:

1. Odd blinking, teeth or hair edges, or lighting that does not match.
2. Audio that sounds flat or has unnatural breathing and pacing.
3. Mismatch between lips and sound.
4. Backgrounds that warp or details like jewelry that change between frames.

Better habits than eye-checking

• Check the source: who first posted it, and do reputable outlets confirm it?
• Search for the original footage or earlier versions.
• For unexpected money or access requests, verify through a separate channel such as calling a known number.
• Be wary of content that provokes strong emotion and urges quick sharing.

Technical and policy responses

What platforms and publishers are doing

Many platforms have introduced rules on manipulated media, ranging from labels to removal of the most harmful content. Newsrooms increasingly verify footage by contacting original sources, checking metadata and consulting forensic specialists. Industry groups have proposed content credentials, a way of attaching a record of how an image or video was captured and edited. Adoption is uneven, and credentials can be stripped, so they help most when combined with the verification habits described above.

Talking to young people about synthetic media

Teenagers encounter manipulated images and videos routinely, and some experiment with the tools themselves. Parents and teachers can explain that creating sexual or humiliating fakes of classmates is harmful and may be illegal, that consent applies to likeness as well as to photos, and that anyone targeted should save evidence and seek help from a trusted adult or the platform. Open conversation works better than prohibition, because the tools are easy to find.

Researchers develop detection tools, and some groups promote provenance standards that attach information about how content was made. Several jurisdictions have introduced or are considering laws on specific harms. None of these is a complete solution, which is why critical media habits remain essential.