Deepfake

A deepfake is synthetic or altered media generated using artificial intelligence, often depicting people saying or doing things that never happened. It can appear as video, audio, or images.

Generative adversarial network: A machine-learning architecture in which a generator creates samples and a discriminator learns to distinguish them from real data. Early face-swapping deepfakes often used adversarial training to produce realistic synthetic faces.
Film dubbing: The replacement of dialogue in a film with speech recorded in another language or performance. Synthetic lip movement can align an actor's visible speech with translated dialogue.
Machine learning: A branch of computing in which systems learn patterns from data to make predictions or generate outputs. Deepfake generators learn visual or vocal patterns from examples rather than following hand-written rules alone.
Photo manipulation: The alteration of a photograph through editing, compositing, or other image-processing techniques. Traditional edits can fabricate images without the AI-generated identity or motion associated with deepfakes.
Deepfake detection: The use of computational or investigative methods to determine whether media has been artificially manipulated or generated. Detection responds directly to the difficulty of distinguishing convincing deepfakes from authentic recordings.
Autoencoder: A neural network trained to compress data into a representation and reconstruct it from that representation. Face-swapping systems can use shared encoders and separate decoders to transfer facial identity.
Digital human: A computer-generated human character designed to communicate or interact through visual or spoken behavior. Deepfake methods can animate digital humans or make them resemble real people.
Neural network: A computational model made of interconnected units that transform input data through learned parameters. Neural networks form the basis of most modern deepfake generation systems.
Computer-generated imagery: Visual content created or rendered with computer graphics techniques. CGI creates synthetic visuals, while deepfakes characteristically alter or imitate identifiable people or recordings.
Content provenance: Information documenting how digital content was created, edited, and distributed. Provenance can establish a recording's history when visual inspection alone cannot establish authenticity.

Linked from 21 pages