A deepfake, a blend of "deep learning" and "fake," is synthetic media in which a real person's face, voice, or movements are generated or grafted onto other footage using neural networks. The term originated in online communities around 2017, and advances in generative models have since made such content far easier to produce.
Early deepfakes relied mainly on GANs and autoencoder-based face swapping; today diffusion models and speech synthesis systems are commonly used to generate video, images, and audio together. Deepfakes have legitimate uses in filmmaking and dubbing, but their growing use in fraud, disinformation, identity theft, and attacks on remote identity verification has made them a focus of detection research and regulation.