How are realistic "deepfake" videos created using AI?

A deepfake video is like when someone changes the face of a person in a movie to make it look like they’re saying something new, but using AI instead of special effects.

Imagine you have a friend who loves to draw. They can take a picture of your face and then draw you smiling, frowning, or even talking. That’s kind of what AI does, but way faster and with videos!

First, the AI learns how someone's face moves when they talk, like watching them say “hello” many times. It studies all the little changes in their eyes, mouth, and eyebrows.

Then, the AI takes another picture or video of that same person doing something else, maybe saying “goodbye.” Using what it learned, it changes the old video to look like they’re actually saying “goodbye,” even if they never did before!

Think of the person’s face as a puppet, and the AI is the clever puppeteer. It knows exactly how to move each string (or muscle) so that the puppet looks real when it talks, just like you or your friend would!