How do 'deepfake' videos manipulate images and voices so realistically?

Deepfake videos make it look like someone is saying or doing something they never actually did, almost as if it happened for real.

Imagine you have a robot that can copy your voice and face perfectly, like a smart mirror in a toy store. This robot uses special computer tricks to take pictures of the person’s face and record their voice. Then, it puts those copied parts into another video, kind of like switching outfits in a dress-up game.

The robot looks at lots of pictures of the person's face from different angles. It learns how each part moves when they talk or smile. Then, it copies that movement onto someone else’s face in the new video, so it looks just like them talking or laughing, even if they never said those words!

It also listens to the person's voice many times and records all the little sounds and tones they use when speaking. The robot then changes another person's voice to match that one, making it sound exactly like the first person is saying the new words.

It’s a bit like having a super-duper copy machine for both faces and voices, but instead of paper, it works with videos!