A deepfake is a piece of synthetic media — video, audio, or a still image — in which a person's likeness, voice, or both have been fabricated or replaced using deep learning techniques. Early deepfakes relied on GANs trained to swap faces frame by frame; more recent systems use Diffusion Models and neural audio codecs that can clone a voice from a few seconds of reference audio. The term combines "deep learning" and "fake."

Think of it like a ventriloquist act where the dummy has been replaced by a photorealistic digital replica of a real person. The voice and movements look convincingly human, but the words being spoken were never actually said by that person.

A typical face-swap pipeline involves three stages: a detection model identifies and crops the face region in each frame, a swap model replaces it with a synthesized version of the target identity, and a blending step feathers the edges to hide seams. Commercial tools like DeepFaceLab and open-source Diffusion-based approaches have made this accessible without specialist hardware.

The technology overlaps closely with GAN and Diffusion research, both of which are used for high-fidelity synthesis. Multimodal models that jointly handle video and audio make voice-synced fakes increasingly convincing. Countering deepfakes is its own active research area, generally called deepfake detection, which applies the same Neural Network techniques to classify media as real or manipulated.

You will encounter the term in journalism and policy discussions about synthetic media regulation, as well as in AI research papers on generative video and audio models.