What is a Deepfake?

A deepfake can copy a face, a voice, or a full video convincingly enough to fool the people who know the target best. Deepfake technology has moved from internet novelty to a working tool for fraud, disinformation, and abuse.

The same models that open real creative possibilities also give attackers a cheap way to fake reality. That's where the real concerns for businesses and individuals begin.

What is a deepfake?

A deepfake is a piece of synthetic media created with AI models trained through deep learning. Put simply, it's an image, video, or audio file that looks authentic but isn't. In many cases a deepfake alters existing content, swapping faces, landmarks, or other objects for AI-generated substitutes. In others, it's built from scratch to show events or conversations that never happened.

Deep learning and artificial intelligence are what separate a deepfake from an ordinary edit. If an image, video, or audio file is changed through conventional methods, like Photoshop without its AI features or traditional Hollywood CGI, it isn't a deepfake.

Deepfakes rely on systems modeled loosely on the neural structure of the human brain to find patterns in data. Building one usually means feeding thousands of data points, such as images, video, and audio samples, into the deep learning network so it learns to reconstruct visual and audio patterns.

Different deepfakes use different underlying AI. Most are produced by algorithms that study the thing they're copying, like the appearance and motion of a face in images or video, or the fine details of a person's voice. Creators then map that onto existing images and footage, or generate entirely new media from it.

The dangers of deepfakes

Deepfakes have legitimate and even beneficial uses, but the technology is dangerous in the wrong hands. An overwhelming majority of deepfake content online is deepfake pornography.

Deepfakes are also used to spread disinformation, erode public trust, and get past security measures, clearing a path for fraudsters and cybercriminals to attack individuals, businesses, the media, and public institutions.

What are image-based deepfakes?

Image deepfakes are synthetically generated images, including photographs and other formats, created or manipulated with AI through deep learning.

Photographs carry weight because we use them to document what's real. A deepfake that can alter an image to show anything its creator wants is a direct threat to that trust. The creative uses are real, but so is the misuse. Deepfake images can spread disinformation, defame public figures, interfere with elections, enable fraudulent advertising and phishing, and steal from artists.

What are video deepfakes?

Deepfake video is synthetically generated video created or altered with AI through deep learning. Making one usually means training a deep neural network on a large set of videos and images of the target. The model learns their features, expressions, and mannerisms, then generates new footage that looks authentic.

Video gets the most attention because it's the media most people consume. Video deepfakes spread disinformation, mislead the public, erode trust in institutions, and malign individuals. As with deepfakes overall, the largest share online is pornography, an abuse of generative AI that violates consent and privacy and drives real harm against women.

Video deepfakes have also been used to manipulate elections, impersonate public figures, and shift opinion on major issues. Cybercriminals use them to impersonate and blackmail people and to get past the biometric security that many companies rely on.

What is deepfake audio?

Deepfake audio is synthetically generated sound created or altered with AI through deep learning.

The most common form is voice cloning, which makes a person appear to say things they never said. Audio deepfakes already show up in disinformation attacks meant to sway elections, mislead the public, discredit real news, and defame individuals.

Voice cloning is a specific problem for any industry that uses voice to verify identity and grant access to accounts and money, especially banking. Companies that field high call volumes report steady deepfake attacks on their systems through voice cloning.

Audio models train on samples of a target voice. Quality depends on how much audio is available, and the newest public tools can produce convincing speech from about 30 seconds of it.

Deepfakes are now part of how media moves through business, journalism, and daily life. The creative uses are real, and so is the danger when a synthetic file reaches a decision before anyone checks it. That's the case for detecting AI-generated voice, video, and images early, inside the workflows where authenticity decides what happens next. See how Reality Defender detects deepfakes.