What Is a Deepfake Video? A Plain-English Guide
In early 2024, an employee at the engineering firm Arup wired out $25.6 million after a video call where every colleague on screen, including the CFO, was an AI fabrication (CNN, 2024). If you have ever wondered what is a deepfake video and how worried you should be, this guide answers both.
- What Is a Deepfake Video and Where Did the Term Come From?
- The 4 Types of Deepfake Videos
- How Deepfake Videos Are Made (The Short Version)
- Real Deepfake Video Examples That Fooled People
- Why Deepfake Videos Are Dangerous
- How to Spot a Deepfake Video
- How to Verify a Suspicious Video
- FAQ
- Conclusion: Know What a Deepfake Video Is, Then Verify
In early 2024, an employee at the engineering firm Arup wired out $25.6 million after a video call where every colleague on screen, including the CFO, was an AI fabrication (CNN, 2024). If you have ever wondered what is a deepfake video and how worried you should be, this guide answers both.
Direct answer: A deepfake video is a video manipulated or fully generated by artificial intelligence to show a person saying or doing something they never did. Deepfakes use deep learning models, such as GANs and diffusion models, to swap faces, clone voices, and synthesize realistic motion.
I work on deepfake detection for a living, so this guide covers the video side of the problem specifically: the four types of deepfake videos, how they are made at a high level, real cases that fooled people, and how to verify a clip you do not trust.
What Is a Deepfake Video and Where Did the Term Come From?
The word deepfake is a blend of deep learning and fake. Deep learning is the branch of AI built on layered neural networks, and it supplies the realism that makes these videos convincing. That blend is the core of the deepfake meaning, and the term dates to a Reddit user called "deepfakes" who shared AI face-swapped clips in late 2017.
A deepfake video, specifically, is the video form of that idea: moving footage in which AI has replaced a face, rewritten lip movements, animated a body, or generated the entire scene from scratch. Audio is usually faked alongside it with voice cloning, so the person sounds right as well as looks right.
Early deep fake video output was crude. Faces flickered, edges smeared, and blinking looked wrong. By 2026, diffusion-based generators produce face swap video output and fully synthetic clips that routinely pass casual viewing. The technology improved faster than most people's instincts did.
If you want the broader, all-formats explainer covering images and audio too, read our guide to what a deepfake is. This page stays focused on video.
- 2017Face-swap deepfake videos first spread online.
- 2019Consumer face-swap apps bring deepfake video to phones.
- 2022Diffusion models sharply improve AI video realism.
- 2024A deepfaked executive video call costs the firm Arup $25.6M.
- 2025Text-to-video generators make synthetic clips easy to produce at scale.
The 4 Types of Deepfake Videos
Not every deepfake video works the same way. Almost everything you will encounter falls into one of four types.
| Type | What the AI changes | Typical real-world use |
|---|---|---|
| Face swap | Replaces one person's face with another's | Celebrity scam ads, impersonation |
| Lip-sync | Rewrites mouth movements to match new audio | Fake statements by politicians and executives |
| Puppet-master | Animates a person's whole head or body | Fake "live" appearances, reenactment |
| Fully AI-generated | Creates the entire video from a prompt | Fabricated events, synthetic people |
Face-Swap Deepfakes
The classic deepfake video. AI maps a target's face onto someone else's body in existing footage, frame by frame. This is the technique behind most celebrity scam ads, such as the AI version of Tom Hanks that promoted a dental plan without his consent, which Hanks publicly disavowed (BBC, 2023).
Lip-Sync Deepfakes
Here the face stays the same, but AI rewrites the mouth region so the person appears to say new words, usually paired with a cloned voice. The 2018 public service announcement in which director Jordan Peele voiced a fake President Obama used this approach to warn viewers about exactly this threat.
Puppet-Master (Full Body Reenactment) Deepfakes
A source actor's movements drive the target's face and body, like a digital puppet. This is the family of techniques behind real-time deepfakes on live video calls, where an attacker's head movements steer a synthetic face. Our guide to deepfake video call scams covers that attack in depth.
Fully AI-Generated Video (Sora-Class Synthetic Clips)
The newest type contains no original footage at all. Text-to-video diffusion models, the class popularized by OpenAI's Sora, generate entire scenes from a written prompt. Strictly speaking, a clip with no real person in it is AI-generated video rather than a deepfake, but the moment a real person is depicted, it crosses into deepfake territory.
How Deepfake Videos Are Made (The Short Version)
You do not need the math, just three ideas.
Training data. Deepfake AI models learn a person's appearance from photos and video of them. Public figures are easy targets because thousands of images of them exist online.
Two model families. Generative adversarial networks (GANs) pit a generator network against a detector network until the fakes become convincing. Diffusion models, the current generation of deepfake technology, refine pure noise into video that matches a prompt or reference. Autoencoders, the original face-swap workhorse, compress and reconstruct faces between identities.
Rendering and blending. The generated face or scene is blended into the footage frame by frame, with color matching and edge smoothing to hide the seams. Those seams are exactly where detection models look first.
That is deliberately high level. For the full technical walkthrough of what is deepfake technology under the hood, see how deepfake videos are made.
Real Deepfake Video Examples That Fooled People
These deepfake examples are all documented, public cases.
The Obama PSA (2018). BuzzFeed and Jordan Peele released a lip-sync deepfake of President Obama saying things he never said, then revealed the trick. Millions saw how convincing the technology already was.
The Arup video call fraud (2024). A finance employee in Arup's Hong Kong office joined a video call where the CFO and several colleagues were all deepfakes. The employee made 15 transfers totaling about $25.6 million before the fraud was discovered (CNN, 2024).
The Tom Hanks dental ad (2023). A face-swapped Hanks endorsed a dental plan he had never heard of. He warned fans on Instagram that the video was AI and not him (BBC, 2023).
We are building a full library of documented cases at deepfake examples and celebrity deepfakes.
Why Deepfake Videos Are Dangerous
Fraud. Video used to be proof. Now a convincing deepfake video can authorize wire transfers, fake job interviews, and impersonate executives. Deloitte's Center for Financial Services projects that generative AI could drive US fraud losses to $40 billion by 2027 (Deloitte, 2024).
Misinformation. Fabricated clips of politicians and officials spread faster than corrections, especially around elections, a risk flagged in the US Department of Homeland Security's primer on synthetic media (DHS).
Harassment and abuse. Non-consensual deepfake imagery is a serious harm category, overwhelmingly targeting women. We do not detail it here; if you are affected, document everything and report it to the platform and law enforcement.
Reputation damage. A fake clip of you or your company can circulate for hours before any takedown. The legal picture varies by jurisdiction and use; see are deepfakes illegal for the current landscape.
How to Spot a Deepfake Video
A condensed checklist. The full version lives in our guide to how to tell if a video is AI generated.
- Watch the face boundary. Blurring or color shifts along the jawline, hairline, and ears are classic face-swap seams.
- Check the lip sync. Mouth movements that lag or fail to match plosive sounds like "p" and "b" suggest manipulation.
- Look at blinking and eyes. Unnatural blink rates, dead-eyed stares, and reflections that do not match the scene.
- Test the physics. Hair, glasses, jewelry, and hands are still where generators most often glitch.
- Question the source. No original upload, no reputable outlet carrying it, urgent emotional framing: treat it as unverified.
How to Verify a Suspicious Video
Eyeballing has limits, and modern fakes are designed to beat them. The reliable next step is forensic analysis.
Upload the clip to our AI video detector and you get a verdict of Authentic, Likely Synthetic, or Inconclusive, plus a TrustScore from 0 to 100, with high accuracy. It detects output from modern video generators including Sora-class diffusion tools. Files are deleted from primary storage within 60 seconds of analysis unless you opt into retention.
In my own review queue, face-swap clips remain the most common deepfake video type users submit, ahead of fully synthetic clips.
Seen a video you do not trust? Check it free. Free accounts include 50 detections per month, with clips up to 2 minutes per check. No card required.
FAQ
Is making a deepfake video illegal? It depends on the jurisdiction and the use. Parody may be protected speech, while fraud, election interference, and non-consensual imagery are increasingly criminalized. Our guide to whether deepfakes are illegal breaks it down.
Can deepfake videos be detected? Yes. Detection models analyze artifacts that generators leave behind, such as blending seams, temporal flicker, and statistical fingerprints, and return a verdict with a confidence score. No detector is perfect, which is why results pair a verdict with a TrustScore.
What is the difference between a deepfake and an AI-generated video? A deepfake video misrepresents a real, identifiable person. An AI-generated video may be wholly synthetic, with no real person depicted. The harm and the law treat the two differently.
How common are deepfake videos? Documented volumes keep climbing, and fraud projections reflect it: Deloitte projects up to $40 billion in US gen-AI fraud losses by 2027. We track the numbers in deepfake statistics.
Who creates deepfake videos? Everyone from hobbyists and meme makers to political operatives and organized fraud rings. The tooling is cheap and widely available, which is exactly why verification matters.
Conclusion: Know What a Deepfake Video Is, Then Verify
So what is a deepfake video? AI-manipulated or AI-generated footage that makes a real person appear to say or do something they never did, across four types: face swap, lip-sync, puppet-master, and fully synthetic. The fakes will keep improving, and your eyes alone will keep falling behind.
The fix is a habit, not a skill: stay skeptical of unverified clips, and run anything that matters through our AI video detector. Start free with 50 checks a month.