Deepfake

Starter
Generative Media
Communications teams
Policy officers
Human rights practitioners

Also called: deep fake, face swap.

Detail

In five seconds

What it is

A deepfake is an AI-generated or heavily altered image, audio clip or video that convincingly imitates an identifiable real person, usually in a recording or event that did not happen as shown.

In plain words

A deepfake is media made by AI tools to look or sound like a real person or event, even when the recording is invented or heavily changed.

An everyday example

Imagine a small media studio in Panama City preparing a historical exhibition. A technician has many real portrait photos of a famous singer, a separate video of an actor reading lines, and software that can combine them. The software studies the singer’s facial features and the actor’s head movements, then produces a new clip that makes it look as if the singer performed those lines on camera. Visitors who only see the final clip may think the recording is genuine.

Where the story stops working

This analogy makes the process sound like a simple cut-and-paste studio trick. Real deepfake systems usually generate many tiny visual or audio details mathematically, frame by frame or sample by sample, rather than just layering one portrait onto another. It also does not capture voice cloning, fully invented people, or the fact that some deepfakes are detected by analysing artefacts and provenance information, not by noticing obvious editing.

Remember thisA convincing imitation of an identifiable person is not proof that the recorded event occurred. Check provenance, source and context.

This term is contested

The exact boundary between 'deepfake' and the broader term 'synthetic media' is unsettled in public use. Some people use 'deepfake' mainly for face and voice impersonation of real people, while others use it more broadly for many kinds of realistic AI-generated media.

Understand it

Why it matters

Deepfakes matter because people often treat photos, audio and video as strong evidence. If realistic media can be fabricated, then seeing or hearing something is no longer enough on its own. This affects journalism, elections, fraud prevention, public trust, reputations, human rights documentation and crisis response. For non-specialists, the key decision point is not learning to spot every fake by eye. It is learning to pause, check the source, ask whether there is independent confirmation, and use provenance or verification tools where available. Deepfakes can also be used in harmless or creative ways, but the same techniques can support impersonation, harassment or deception.

How a deepfake is made and why it can mislead

After viewing this diagram, the reader should be able to describe the basic flow from source material to generated output and explain why realism does not prove authenticity.

01
1. Source material
Real photos, videos or voice recordings of a person are collected, along with prompts or a separate performance to imitate.
Show a box with small icons for photos, video frames, an audio waveform and a text prompt. Label it 'Real examples + instructions'.
02
2. Pattern learning
The model analyses patterns such as facial structure, lip movement, lighting or vocal style.
Show a box with a simplified network or gear icon. Add short labels around it: 'face', 'motion', 'voice', 'style'.
03
3. Generated media
The system produces new frames or audio that imitate the person or scene, even if the event never happened.
Show a box with a video thumbnail and speaker icon. Add a tag: 'synthetic output'.
04
4. Human response
Viewers may accept the media as real unless they check context, provenance or other evidence.
Show a final box with an eye icon and a checklist icon. Split the box visually into 'seems real' and 'verify source'.

A left-to-right flow diagram with four boxes connected by arrows. It starts with real source material, moves to a generative model, then to a synthetic output, and ends with viewer judgement and verification. Read from left to right to see how believable media can be produced from real examples and why checking the source matters.

Hovering or tapping each box reveals a short note with one example: face swap, voice clone, invented scene, or provenance check.

Describe this without the diagram

This diagram explains the process in four stages. First, real source material is gathered, such as photos, recordings or a script. Second, a generative model analyses patterns in that material, including appearance, movement or voice traits. Third, the model creates synthetic media that imitates a person or event. Fourth, a viewer encounters the result and may mistake it for real unless they verify the source, context or provenance. The main point is that realistic media can be generated from real examples, so appearance alone is not proof of authenticity.

With reduced motion: A static four-box flow chart is shown instead of animation. Arrows connect source material, pattern learning, generated media and human response, with the same labels and example notes visible at once.

A deepfake is a type of synthetic media that uses generative methods to imitate an identifiable real person. Inputs may include photos, video clips or voice recordings of that person, plus text instructions or another performance. The output may make the person appear to say or do something they never said or did.

A common form is a face swap, where one person’s face is placed onto another person’s body in a video. Another form uses voice generation to imitate a person from sample recordings and produce new speech. Voice, image and video generation can create deepfakes when used for this kind of false depiction. Media showing entirely invented people or scenes is broader synthetic media, not automatically a deepfake.

Under the surface, a model uses patterns learned during training and may analyse reference material supplied for the task. It can reproduce details such as facial movement, lighting or changes in a voice across words. Different systems use different methods, including diffusion models, generative adversarial networks and other media-generation systems.

People often confuse deepfakes with all fake or misleading media. A slowed video, cropped photo or false caption can mislead without generative methods. Deepfake describes how media is generated or altered. Misinformation and disinformation concern misleading content and whether it spreads by mistake or on purpose.

How “A forged video portrait in a studio” maps onto the real thing

In the story	In the AI system
Portrait photos of the singer	Training or reference material showing a real person’s appearance
Video of the actor reading lines	Source performance that provides movement, timing or expression
Software combining face and movement	Generative model that creates altered video frames
Final exhibition clip	Deepfake output that may look authentic to viewers

A worked example

A fabricated video appears to show a well-known presenter endorsing an investment scheme in a recording that never took place.

Starting from: Public interview footage, speech samples, a fabricated script and a source video of another person performing the lines.

A generative system extracts statistical patterns in the presenter's facial appearance, movements and vocal timbre from the reference material.
The source performance supplies head movements, expressions and timing for the new video.
Generated facial imagery and a cloned voice are aligned with the script, including lip synchronisation and speech rhythm.
The synthetic elements are blended into the source video, then compressed and circulated as though they came from a genuine recording.

Result: The finished clip imitates an identifiable person but does not document a real endorsement. Its authenticity must be assessed through provenance, the original source, contextual checks and corroborating evidence, not visual realism alone.

In a United Nations workplace

During the annual workplan review, a national officer in a documentation and library service receives a video circulating on staff messaging channels. It appears to show the chief of section announcing that a digitisation activity has been cancelled, and a colleague asks whether its target should be removed from the workplan before clearance at close of business.

The officer should not treat a familiar face, voice and office background as proof. A deepfake can reproduce all three. The original source, file metadata, meeting record and direct confirmation through an established channel matter more than how convincing the clip looks.

Before you try thisDo not upload an internal recording, personal data or security-sensitive media to an unapproved detection service. This may disclose the material without establishing authenticity. Preserve the original file and metadata, use approved reporting channels, and remember that automated deepfake detection is not conclusive on its own.

Where you may hear it

A morning media-monitoring briefing
A communications duty handover after a suspicious clip appears
A security-awareness briefing on impersonation attempts
An inter-agency discussion on election-related misinformation

Avoid confusion

Often confused with

Common misconceptions

Common misconception: Every fake-looking video online is a deepfake.
Correction: Not necessarily. Many misleading clips are simple edits, old footage reused in a new context, slowed audio, false subtitles or manipulated captions. 'Deepfake' usually refers to media created or altered with generative techniques that imitate reality, especially a person’s face or voice.
Common misconception: If a deepfake looks perfect, ordinary people have no way to respond.
Correction: Visual inspection alone can fail, but people still have useful checks. They can look for the original source, compare with trusted reporting, check whether reputable accounts carry the same material, and use provenance or forensic review where available. The right response is verification, not guessing from appearance alone.
Common misconception: Deepfakes are only a celebrity problem.
Correction: Public figures are common targets because there is lots of source material, but ordinary people can also be harmed through voice scams, harassment, non-consensual sexual content or false evidence. The risk is wider than entertainment and affects personal safety, trust and reputation.

Quick check

Which example best fits the term 'deepfake'?

Go deeper

The simple version

A deepfake is media made by AI tools to look or sound like a real person or event, even when the recording is invented or heavily changed.

The technical version

A deepfake is a subset of synthetic media produced with generative methods that imitate identifiable visual or audio characteristics, often of a real person. Inputs may include reference images, video frames, speech samples, text prompts or a source performance. The system learns statistical patterns in facial appearance, motion, lip synchronisation, vocal timbre or scene structure, then generates new pixels or waveforms to create altered or fully synthetic content. Common implementations have included face reenactment, face swapping, voice cloning and text-to-video generation, using model families such as generative adversarial networks or diffusion models. The term is used widely in public discussion, but in practice its boundary with broader synthetic media is blurry, especially as modern generation tools can create realistic output without a traditional face-swap pipeline.

Reviewed external resources

Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This link provides a policy and ethics framework for thinking about AI-generated media and its risks, which suits policy, governance and public-sector readers.
Intermediatelink checked 2026-07-31
Primary sourceContent Credentials (opens in a new tab)Content Authenticity Initiative
This link explains an approach to content provenance and authenticity signals, which suits readers looking for practical ways to assess whether media has trustworthy origin information.
Starterlink checked 2026-07-31
EncyclopaediaDeepfake (opens in a new tab)Wikipedia
This link gives a broad introductory overview of the term, its history and common uses, which suits readers who want a quick starting point.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Deepfake (opens in a new tab)YouTube search
This is a search results page for video explanations from different creators, which suits readers who prefer to compare several introductory video styles rather than rely on one source.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Deepfake (opens in a new tab)Reddit search
This is a search results page for community discussion and examples, which suits readers who want to see public questions, reactions and debates but should be read critically.
search results, not one pagelink checked 2026-07-31

Where this came from

Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.

The one thing to remember

A convincing imitation of an identifiable person is not proof that the recorded event occurred. Check provenance, source and context.