March 2025 · 7 min read · Deepfakes
A deepfake is a synthetic media file — typically a video or audio recording — in which a person's likeness, voice or actions have been digitally replaced or manipulated using artificial intelligence. The term combines "deep learning" (the AI technique used) with "fake." Deepfakes can make people appear to say or do things they never said or did, with a level of realism that makes them extremely difficult to detect without specialised forensic tools.
Modern deepfakes are created using a class of machine learning architecture called autoencoders combined with Generative Adversarial Networks (GANs). The process works by training two competing neural networks — a generator that creates fake content and a discriminator that tries to detect whether content is real or fake. Over thousands of iterations, the generator becomes progressively better at fooling the discriminator, resulting in increasingly realistic output.
To create a face-swap deepfake, the AI needs a large collection of images or video frames of both the source face (the person whose face will be used) and the target face (the person in the original video whose face will be replaced). The encoder learns to extract facial features from both people, and the decoder reconstructs the source person's face in the expressions and lighting conditions of the target video.
Face swap deepfakes replace one person's face with another's in a video. This is the most common type and the one most people are familiar with. Voice deepfakes clone a person's voice using text-to-speech models trained on recordings of that person speaking. Puppet deepfakes animate a still photograph to move, speak and express emotions. Full body deepfakes replace an entire person in a video. Text deepfakes generate written content that mimics a specific person's writing style.
Deepfakes have been used in election interference campaigns, where politicians have been made to appear saying things they never said. In 2024, a deepfake audio of a US presidential candidate was used in voter suppression robocalls. Deepfakes have been used in financial fraud, with criminals using voice clones to impersonate executives and instruct employees to transfer large sums of money. Cybercriminals have used deepfakes to bypass video-based identity verification systems at banks and financial institutions.
The most common use of deepfake technology remains non-consensual intimate imagery — synthetic explicit content created using a real person's face without their consent. This form of abuse disproportionately targets women and has devastating psychological consequences for victims.
Visual inspection can reveal some deepfakes, particularly older or lower-quality ones. Look for unnatural blinking patterns, facial edges that blur or shimmer when the person moves, inconsistent lighting between the face and background, and lip sync that does not perfectly match the spoken audio. However, high-quality modern deepfakes can fool visual inspection entirely, making automated forensic analysis essential.
Forensic detection tools like Chicken AI analyse dozens of signal dimensions that are invisible to the naked eye — including temporal consistency between frames, facial landmark drift, compression artefact patterns and audio-visual synchronisation — to determine whether a video has been manipulated.
Upload any video or image and our ChickenMind™ Engine will analyse it for deepfake indicators in under 3 seconds.
Analyse a Video Now →