A deepfake is a piece of media — an image, video, or audio recording — that has been created or altered using artificial intelligence to depict something that did not actually happen. The term, a portmanteau of "deep learning" and "fake," emerged in the late 2010s and has since become synonymous with the broader challenge of synthetic media. Deepfakes can swap faces in videos, generate photorealistic images of people who do not exist, clone voices from short audio samples, and manipulate the words and actions of real people in footage. As the technology has become more accessible and convincing, it has emerged as a significant threat to online trust, identity verification, and the integrity of visual evidence. For the technology that fights back, see our complete guide to deepfake detection.
What Is a Deepfake
At its core, a deepfake is the product of generative AI models — typically generative adversarial networks (GANs) or diffusion models — that learn to synthesize realistic media by training on large datasets of real images, videos, or audio. The most common deepfake technique is face swapping, where the face of one person is replaced with the face of another in a video, while preserving the original expressions, lighting, and movements. Another prominent application is fully synthetic face generation, where AI creates a photorealistic image of a person who does not exist — these "deepfake faces" are increasingly used to populate fake social media profiles, dating accounts, and scam operations. Voice cloning, a related technology, can replicate a person's voice from just a few seconds of audio, enabling fraudulent phone calls and messages. The defining characteristic of a deepfake is that it is designed to be mistaken for genuine media.
How Deepfakes Work
Face-swap deepfakes typically use an encoder-decoder architecture. The encoder learns to compress a face into a shared latent representation that captures expression, pose, and lighting. Separate decoders are then trained to reconstruct each individual's face from that representation. At inference time, the encoder processes the source face, and the target person's decoder reconstructs the face — effectively swapping identities while preserving the original movement and expression. Fully synthetic faces are generated by models like StyleGAN or diffusion-based systems, which learn the distribution of real faces and can sample new, photorealistic faces from that distribution. The quality of modern deepfakes is high enough that many are indistinguishable from real media to the human eye, particularly on small screens or in low-resolution contexts. For the technical countermeasures, read our guide on how deepfake detection works.
The Threat of Deepfakes to Identity Verification
Deepfakes pose a direct threat to identity verification, which increasingly relies on photographs and video. A scammer can use a synthetic face to create a convincing dating profile, a fake social media account, or a fraudulent business persona — none of which can be traced to a real person. This undermines the effectiveness of face search, because the "face" being searched may not correspond to any real individual. Deepfakes also threaten video-based verification: a scammer can use real-time face swapping to pass a video call, making it appear that they are the person in the profile photo when they are not. The rise of deepfakes means that a single verification method is no longer sufficient. Robust identity verification now requires layering multiple checks — face search to find where a photo appears publicly, deepfake detection to determine if the photo is synthetic, and behavioral or contextual signals to corroborate identity.
Where Deepfakes Cause the Most Harm
- Romance and financial scams — synthetic faces power fake dating profiles that are impossible to trace to a real person.
- Political disinformation — deepfake videos of public figures can spread false statements and damage reputations.
- Corporate fraud — voice clones of executives have been used to authorize fraudulent wire transfers.
- Non-consensual intimate imagery — real people's faces are placed in explicit content without their consent.
- Synthetic identities — entirely fabricated personas are used to open accounts, pass KYC, and commit fraud at scale.
How to Detect a Deepfake
Detecting deepfakes is an arms race: as detection methods improve, generation methods evolve to evade them. However, several signals remain useful. Visual artifacts — unnatural blinking patterns, inconsistent lighting, blurry boundaries around the face, mismatched skin tones, and artifacts in teeth or hair — can betray a deepfake, particularly in lower-quality outputs. Inconsistencies between audio and video, such as lip movements that do not perfectly match speech, are another indicator. Metadata analysis can reveal whether a file has been processed or edited. Specialized deepfake detection tools use machine learning models trained to identify the subtle fingerprints left by generative algorithms, such as frequency-domain artifacts and inconsistencies in pixel patterns. For identity verification, the most robust approach combines deepfake detection with face search: if a photo returns zero public matches across the entire web, that is a strong signal that the face may be synthetic. To learn more, read our guide to detecting AI-generated faces.
Deepfakes and the Future of Face Search
The rise of deepfakes is reshaping how face search engines operate. A face search that returns no results is no longer simply a sign of a private person — it may indicate a synthetic face. This makes the absence of results a meaningful signal in itself, one that should prompt additional scrutiny. face search engines like facesearching are increasingly integrated with deepfake detection, so that users receive not only a list of where a face appears publicly but also an assessment of whether the face is likely to be real or AI-generated. As generative models continue to improve, the combination of face search and deepfake detection will become an essential part of any identity verification workflow. To run a search and check a photo, visit the facesearching home page.
Deepfakes blur the line between real and synthetic — the future of identity verification depends on pairing face search with deepfake detection, so that an absence of public matches becomes a warning rather than a dead end.