What Is a Deepfake Video? Deepfakes, Deep Learning, and Warning Signs

Here's the finding that stopped researchers in their tracks: when scientists studied who was likely to create deepfake pornography, fake explicit videos made by pasting someone's real face onto another body without their consent, they looked for the obvious red flags. Narcissism. Dark personality traits. The psychological profile of someone you'd recognize as dangerous. And they found... nothing. No connection. Zero link between those personality measures and willingness to create this kind of abuse.

What did predict it? Attitudes. Ordinary, learned, changeable beliefs about women, entitlement, and what counts as "just a joke."

The Research That Changes How We Think About This

What Is a Deepfake Video and Why Does Deepfake Technology Matter Here

Before going further, it helps to define the term at the center of this research. A deepfake video is video content where someone's face, voice, or likeness has been digitally swapped or synthesized so that it looks like they said or did something they never actually said or did. Deepfake technology relies on machine learning models, often a type of deep learning system, trained on real photos and video of a person until the software can convincingly generate new, fake footage of them. That same underlying technology is what makes deepfake videos usable for entertainment, satire, and fraud, and it's also what makes deepfake pornography possible at a scale that simply wasn't achievable by hand a decade ago.

The study came out of Edith Cowan University in Australia, published in the Journal of Sexual Aggression. Researchers surveyed 213 heterosexual men, asking them about their personality traits, their attitudes toward sexual harm, and, through carefully designed fictional scenarios, their willingness to create deepfake pornographic content.

The design was deliberate. They separated personality measurement from attitude measurement on purpose, so they could see which one actually drove behavior. The answer was clear: personality traits didn't move the needle. Attitudes did.

Specifically, the attitudes that predicted willingness to create this content were ones that minimize or excuse sexual harm, things like believing women exaggerate abuse, that certain victims "brought it on themselves," or that sharing explicit images without consent is "just a laugh." Researchers call this cluster of beliefs "rape myth acceptance", meaning false beliefs that trivialize or excuse sexual violence. It sounds like academic language, but the underlying ideas are things you've probably heard at a dinner table or in a comment section. This article is part of a series, start with 1 In 3 Teens Now Hit By Fake Ai Nudes Heres What To Do Tonig.

That number, 464% growth in a single year, matters here. Because if deepfake abuse were driven by a fixed, rare personality type, that kind of explosive growth would make no sense. Dangerous personalities don't multiply that fast. Normalized attitudes do. When a behavior becomes easier to do AND easier to excuse, more ordinary people do it. That's exactly what happened.

---

The "Social Distance" Effect, Why Celebrities Feel Like Fair Game

Deepfakes, Synthetic Media, and Security: Why Information Spreads Faster Than Facts

Deepfakes belong to a broader category called synthetic media, meaning any image, audio, or video that has been generated or altered by AI rather than captured directly from reality. Synthetic media isn't automatically harmful; some of it is clearly labeled satire, film effects, or research. But when synthetic media is passed off as real, false information spreads fast, and by the time accurate information catches up, the damage to a person's reputation is often already done. This is also a security problem, not just a privacy one, because the same neural network techniques used to generate deepfake videos can be pointed at voice cloning, fake identity documents, or impersonation attempts designed to defeat security checks. Understanding the security angle helps explain why researchers, platforms, and lawmakers increasingly treat deepfakes as an information integrity issue and not only a personal one.

One of the study's most revealing findings involves who people were more willing to target. Participants reported significantly greater willingness to create deepfake pornography when the target was a celebrity versus someone they personally knew.

Think about what that tells us. It's not about the harm being different, a fake explicit video destroys someone's life whether they're famous or not. The harm is identical. But the psychological distance between creator and target changes how people feel about doing it.

When you don't know someone personally, it's easier to think of their face as a kind of raw material. Something you found online, something that's "already out there." The celebrity's face becomes almost abstract, a data point, not a person. That mental shift is what researchers call a permission structure: a set of mental conditions that make a harmful act feel acceptable, or at least not-quite-real.

"Attitudes minimizing or excusing harmful behavior significantly predicted willingness to create deepfake pornography, while no links were found between dark personality traits such as narcissism and the likelihood to create such content." Marns & Hollett (2026), as reported by Phys.org

This also explains part of why 99% of deepfake pornography targets women. That's not a coincidence of who has famous faces. It reflects something systemic: a broader cultural attitude about whose image is available for consumption, whose body is treated as content, and whose discomfort is treated as optional. The pattern is structural, not individual. You can't explain a 99% gender skew with individual psychology. You explain it with shared cultural beliefs.

---

Why We Keep Looking for the "Obviously Bad Person"

Here's where most people get this wrong, and honestly, it's an understandable mistake. Previously in this series: Your Face Just Became A Password You Can Never Change.

We're wired to protect ourselves by spotting dangerous individuals. The creepy stranger. The obvious red flag. The person who just feels off. That instinct served humans well for thousands of years, and it's deeply baked in. So when we hear about something as violating as deepfake sexual abuse, our brain immediately reaches for the same toolkit: who is the obviously disturbed person doing this?

The research says: that's not the right question. No links were found between dark personality traits, narcissism, psychopathy, that whole category, and the likelihood to create deepfake pornography. The person doing this doesn't have a sign on their forehead. They might be the guy who laughs a little too hard at certain jokes. The one who says "it's not that serious" when someone complains about harassment. The one who thinks sharing a fake image is "just trolling."

That's the scarier lesson. Not that monsters are hiding among us, but that ordinary people, holding ordinary-seeming attitudes, are committing serious harm and mostly not recognizing it as harm at all. When Edith Cowan University researchers looked at why people self-reported creating image-based sexual abuse, the justifications ranged from "sharing for fun" to chasing social approval, responses that showed almost no awareness that they'd done something harmful. That's not a monster's rationalizing. That's a permission structure doing its job.

---

What This Actually Changes, For Parents, Schools, and Anyone Paying Attention

Think of it this way. A harmful personality type is like a reckless driver, someone inherently dangerous behind the wheel. But harmful attitudes are like a road with no speed limit signs, surrounded by other drivers who treat speeding as completely normal. The second scenario doesn't require reckless people. It just requires a culture that removes the guardrails. Remove the social friction that would otherwise make someone stop and think "wait, this is wrong", and ordinary people do harmful things.

That reframe matters enormously for anyone trying to protect young people. Teaching kids to "spot the dangerous person" is useful, but it's incomplete. Because the research says the dangerous person might look perfectly normal, might even be popular, funny, well-liked. What's less normal, and what we can actually teach people to notice, are the attitudes.

Attitudes that treat women's images as public property. Jokes where the punchline is someone's humiliation. Group chats where sharing explicit content without consent gets social upvotes instead of pushback. Online spaces where "it's just a meme" or "she's famous, she can handle it" functions as a permission slip. Those are the warning signs the research points to, and they're visible, if you know what you're looking at. Up next: Government Login Identity Verification Malta What It Means F.

At CaraComp, we work in facial recognition and identity verification, the exact technology that makes deepfake creation possible at scale. Understanding why people misuse it isn't just academic. It shapes what protections matter most: not just detecting fake images after they're created, but building the kind of awareness that stops the justifications before anyone opens the software.

The peer-reviewed study by Marns & Hollett points toward a hopeful conclusion buried inside a difficult finding. Personality traits are largely fixed, you can't teach someone out of narcissism. But attitudes? Attitudes are learned. Which means they can be unlearned. They can be challenged in a group chat, called out in a classroom, named for what they are at a dinner table.

So here's the question worth sitting with: if the warning sign isn't "this person seems dangerous" but "this person thinks humiliating someone with their image is funny", what would it look like to actually call that out, the next time you see it? Not after the video exists. Before anyone decides to make it.

The research didn't just identify a problem. It handed us a different place to intervene. That's the aha moment: the prevention work isn't about detecting monsters. It's about making sure the attitudes that create permission structures don't go unchallenged in the ordinary, everyday places where they quietly take root.

So what is a deepfake video, in plain terms, once you strip away the technical jargon? It is footage engineered to make a real person appear to say or do something they never did, built using deep learning models trained on that person's actual face and voice. Deepfake videos range from harmless entertainment, like a face-swapped movie clip, to serious fake videos used for harassment, fraud, or nonconsensual pornography. The research covered here focuses specifically on that harmful end of the spectrum, where deepfake technology is used without consent to humiliate or exploit real people.

It's worth separating deepfake video from the broader idea of fake videos in general. Not every fake video uses a neural network or deep learning; some fake videos are just clever editing, staged footage, or misleading captions on real, unaltered clips. What makes a deepfake video distinct is the use of AI models trained on real images and voice recordings of a specific person to synthesize new, false footage that looks and sounds like them. That distinction matters for security teams and platforms trying to detect deepfake videos, because the detection methods differ depending on whether the manipulation involves synthetic media generation or simple editing tricks.

Voice is often overlooked in conversations about deepfake video, but it deserves attention. Many deepfake videos pair a synthesized face with a cloned voice, so the video and the audio reinforce each other and feel more convincing together than either would alone. Scammers have used voice cloning on its own, without any video at all, to impersonate a family member or executive over the phone. When voice cloning and deepfake video techniques are combined, the resulting content can be extremely difficult for an untrained viewer to identify as fake, which is one more reason security awareness has to include audio, not just images.

False information built on deepfake videos travels differently than an ordinary rumor. Because the fake video looks real, people share it before checking whether the underlying information is accurate, and corrections rarely reach as many people as the original false information did. That's true whether the deepfake video is being used to embarrass a private individual, impersonate a public figure, or manufacture a fake news clip. Learning to pause before sharing any startling video, and asking where it actually came from, is a small habit that meaningfully slows the spread of false information.

Real protection against deepfake videos combines two things: better detection technology and better human judgment. On the technology side, researchers are building tools that flag inconsistencies a neural network leaves behind, subtle blending errors around the jaw or eyes, unnatural blinking patterns, or audio that doesn't quite sync with lip movement. On the human side, learning what a deepfake video actually is, and staying skeptical of shocking footage until it's verified, remains one of the most reliable defenses available to ordinary internet users today.

Schools, workplaces, and families all benefit from the same basic media literacy: treat a shocking video as a claim to verify, not a fact to forward. Ask who posted it, whether a credible outlet has confirmed it, and whether the video's story feels engineered to provoke an emotional reaction before you've had time to think. Deepfake technology will keep improving, and fake videos will keep getting harder to spot on sight, so the habit of pausing before sharing matters more than any single detection tool.

It helps to walk through a few concrete examples of deepfakes so the term stops feeling abstract. A deepfake might be a video of a politician appearing to say something they never said, a celebrity's face pasted into a movie scene they never filmed, or a private citizen's photo turned into explicit content without consent. Each of these deepfakes uses the same underlying deep learning approach, even though the intent behind them ranges from harmless parody to serious abuse. Recognizing that deepfakes come in many forms, not just one obvious type, is the first step toward spotting them in the wild.

Deep learning is the branch of machine learning that makes modern deepfakes possible, and understanding it in plain terms helps demystify the whole topic. A deep learning system studies thousands of images or video frames of a real person's face until it learns the patterns of how that face moves, blinks, and forms expressions. Once trained, the same deep learning model can generate new footage that follows those learned patterns, which is how a deepfake video ends up looking so convincingly like the real person. This is also why deep learning needs lots of source material, a person with many public photos and videos is, unfortunately, an easier target for this kind of synthetic media.

The internet has made both the creation and the spread of deepfakes faster than ever. A deepfake video can be produced on a home computer and uploaded to the internet within hours, reaching thousands of viewers before anyone has a chance to verify whether the footage is real. Because so much information now travels across the internet without any built in fact-check, a single convincing deepfake can outrun the truth long before corrections catch up. That speed is part of why researchers treat deepfakes as an internet-scale information problem, not an isolated incident.

Not all synthetic media is created with harmful intent, and it helps to hold that nuance alongside the warnings in this article. Filmmakers use synthetic media for de-aging actors or recreating historical figures respectfully, and researchers use synthetic media to test detection tools without exposing real victims. The difference between acceptable synthetic media and harmful deepfakes almost always comes down to consent and disclosure, whether the person depicted agreed to it and whether the audience is told the footage isn't authentic. Keeping that distinction in mind helps people react proportionately instead of treating every instance of synthetic media as equally alarming.

Learning to spot a deepfake video is a skill that improves with practice, much like learning to spot a photoshopped image once required. Look closely at how light falls across a face compared to the rest of a video, whether shadows on the neck and jaw match, and whether the edges around hair or glasses look slightly blurred or warped. Learning these small tells will not catch every fake video, especially as deep learning tools improve, but it raises the bar for what fools you and buys time to verify suspicious footage before sharing it further.

Fake videos are not new, but the internet has changed how fast a fake video can travel and how convincing it can look once it arrives. Before deep learning made deepfake videos possible, a fake video usually relied on selective editing or an out of context clip labeled as something it wasn't. Today's fake videos can combine that old trick with an AI generated face or voice, which means false information hiding inside a fake video is harder to catch just by watching once. That combination is exactly why media literacy and detection technology need to improve together, not separately.