A lip sync deepfake modifies the mouth movement of a real video so it appears to match new audio. The original footage is otherwise authentic, which makes lip sync attacks particularly hard to detect at a glance.

• New audio is generated, often using cloned voice of the target
• A model regenerates the mouth and lower face region of the source video to match the new audio’s phonemes
• The rest of the frame stays unchanged, preserving authentic body language and surroundings

• Doctored political speeches with synthetic statements grafted onto real footage
• Fake celebrity endorsements using genuine appearance footage
• Manipulated press conference clips spread as authentic news
• Defamation campaigns putting fabricated quotes into real interviews
• Content used to discredit real reporting by inserting compromising statements