Plain-English explanation
Deepfake technology uses machine learning, including face swapping and generative video models, to place a real person's likeness into footage, audio, or images in which they did not participate. The result can range from obviously fake to highly convincing, depending on the source material and technique quality.
A deepfake depicts a real, identifiable person's likeness doing or saying something that didn't happen. Content generated entirely from an AI's own imagination (an original character with no real-world counterpart) is not a deepfake, even though it's produced with similar underlying technology. Permission for that identity-linked use is a separate question covered by likeness consent.
Legitimate and harmful uses
Deepfake and face-synthesis techniques have real, legitimate applications: film and TV visual effects (de-aging or stand-in work with an actor's own consent and involvement), dubbing and lip-sync for licensed foreign-language releases, restoring or animating historical footage, satire and commentary protected as free expression in many jurisdictions, and academic research into detection and media forensics.
The same underlying technology is also used to generate sexual content of real people without their consent, taking someone's face or likeness and placing it into explicit material they never agreed to appear in. This is categorically different from generating sexual content of an original AI character, which involves no real person's identity at all.
Generating non-consensual sexual deepfakes of real people is harmful (it causes documented psychological, reputational, and in some cases professional harm to the person depicted) and it is increasingly illegal. In the United States, the federal TAKE IT DOWN Act (signed May 2025) criminalizes knowingly publishing non-consensual intimate imagery, including AI-generated deepfakes, and most U.S. states now have their own deepfake-specific laws. This page discusses the concept and its implications; it is not a guide to creating deepfakes, and no such instructions are provided here or elsewhere on this site.
Adult-AI use
The word "deepfake" is often used loosely to describe any AI-generated adult content, but that's imprecise and matters for this site specifically: most AI companion, character-chatbot, and adult image/video generation activity involves entirely original, invented characters with no real-person basis: that's synthetic media, not deepfakes. Reputable adult AI platforms generally prohibit generating sexual content of real, identifiable people specifically because of the deepfake/non-consent problem, while permitting generation of original characters. Understanding the distinction matters both for accurately describing what a given tool does and for understanding why platform policies treat "generate an original character" and "generate a real person's likeness" as fundamentally different requests.
How it differs from synthetic media
Every deepfake is synthetic media, but not every piece of synthetic media is a deepfake. Synthetic media is the umbrella category for anything AI-generated or AI-modified; deepfake is the narrower subset specifically involving a real person's likeness depicted doing or saying something they didn't. An AI-generated image of a wholly invented character is synthetic media but not a deepfake, because there's no real identity being misrepresented.
Limitations and misconceptions
- Not all synthetic or AI-generated content is a deepfake: the term specifically requires a real, identifiable person's likeness.
- Deepfake quality varies enormously; some are easily spotted, others are convincing enough to fool casual viewers, and quality has generally improved as underlying generation models have improved.
- Detection is an active and imperfect research area: no detection method is claimed to catch every deepfake reliably, and detection tools themselves can be defeated as generation techniques improve.