How to Deepfake Video: Make Deepfake Technology, Faceswap Guide

Only 0.1% of peopleacross every age group tested, correctly identified every deepfake they were shown. Not 10%. Not 1%. Zero-point-one. And the group that was most overconfident about their ability to spot fakes? Young adults. The very people we've been counting on to save us from this mess.

Here's what's actually going on. We've spent two years asking the wrong question. Everyone, news articles, school curricula, your savvy nephew, has been asking: "Can you tell it's fake?" That question made sense when deepfakes had glitchy hands, blurry teeth, and ears that didn't quite match. Those days are gone. The question that matters now is: "Can you verify it before you do something you can't undo?"

That shift, from spotting to verifying, is the whole ballgame. And most of us haven't made it yet.

---

The Gut-Check That's Letting You Down

Why Understanding How People Make Deepfake Videos Changes What You Fear

Your gut was built for detecting lies told by people standing in front of you. Micro-expressions, voice cracks, a shifty glance. It's actually pretty good at that, honed over thousands of years of human interaction. But a video generated by an AI model that has processed millions of hours of real human footage? Your gut never evolved for that. It doesn't stand a chance.

The Fulcrum ran a piece this week that gets at a genuinely interesting question: does Gen Z's natural skepticism about AI-generated content give them a democratic advantage at the polls? The finding is more complicated, and more unsettling, than a simple yes. This article is part of a series, start with How Deepfake Video Detection Actually Works.

Younger voters do carry a useful default suspicion. They grew up with filters, face-swapping apps, and viral "is this real?" posts. That suspicion is protective. But suspicion isn't the same as skill. When researchers actually tested detection ability, showing people real and fake videos side by side, the age advantage nearly disappeared. And young adults were the most likely to say they were confident in their answers while being wrong.

That number should stop you cold. It stopped me. We are collectively walking around believing we can eyeball our way to safety, and almost none of us actually can.

---

The Election Problem Is Really an Everyone Problem

How Deepfake Technology Reached the Point Where Anyone Can Make a Deepfake

Political deepfakes are where this gets easy to see, but hard to fix. In a Kentucky race this cycle, campaign operatives were explicit about the math: older voters make up the bulk of the electorate, and older voters are less likely to clock an AI-generated video as fake. The targeting wasn't accidental. As one account of the strategy put it, those behind fake political ads were banking on the fact that "the older generation won't realize it's an AI-generated lie."

Meanwhile, in a New York race, younger voters immediately identified AI-generated attack ads and mocked them online, turning the content into a liability for whoever deployed it. Same technology. Completely different outcome, based entirely on audience.

That gap is a real problem. According to data on the 2022 midterms, about 27.6% of 18-to-24-year-olds voted, compared to 68% of voters aged 65 to 74. The generation with the most instinctive skepticism toward AI content has the smallest footprint on actual election results. The generation most vulnerable to convincing fakes holds far more electoral weight. That's not a techie concern, that's a democracy concern.

And roughly 30 states now have some form of AI disclosure law on the books, but The American Prospect found that explicit AI disclaimers simply didn't appear in multiple political ads that should have had them. Laws exist. Enforcement is another matter entirely. Previously in this series: That Panicked Call From Your Kid 3 Seconds Of Tiktok Is All .

In the UK, Demos polled adults ahead of the May 2026 elections and found that 30% had seen political deepfakes online in the preceding month, with 16% encountering them more than five times. That's not a fringe event. That's saturation. The UK Electoral Commission launched a deepfake detection pilot in April 2026 specifically to monitor online content for AI-generated audio and video designed to mislead voters. It's a real program, and the fact that it exists tells you everything about how serious this has gotten.

---

So What Actually Protects You?

What Deepfake Apps Can Do That Your Eyes Cannot Catch

Not your eyes. Not your ears. Not your gut. What protects you is a habit, a small, specific pause before you act on anything surprising or emotionally charged that comes through your screen.

The habit is this: verify through a second channel before you react. That means if you get a voice message from your daughter sounding panicked and asking for money, you hang up and call her back on the number you already have saved. If you see a video of a politician saying something shocking, you check whether any outlet you actually trust has covered it, not just whether it looks real. If a friend sends you a clip that makes your blood boil and you want to share it immediately, that urgency is exactly the moment to wait.

"The best way to spot deepfakes is to examine context, not appearance, and verify through another trusted channel before acting." Analysis from The Fulcrum

This isn't complicated. It doesn't require an app or a subscription or a tech degree. It just requires interrupting the automatic reflex to react firstand that reflex is exactly what bad actors are counting on. Anger, fear, and disbelief all make you faster and sloppier. Fake content is specifically engineered to trigger those exact feelings.

The World Economic Forum's 2026 analysis of AI disinformation describes this as a shift from a detection problem to a trust-infrastructure problem. In plain language: the question is no longer "is this fake?" but rather "what process do I use before I decide to believe and act?" That's a meaningful reframe. Detection is a skill. A process is a habit. Habits are more reliable.

If you've ever had that moment of doubt, looking at a photo or a video and thinking, something feels off, but I can't quite say whatthat instinct is worth honoring. The problem is that as the technology improves, that feeling of "something's off" is going to come less and less often. The fakes are getting smooth. The tells are disappearing. Which means the pause, the deliberate second-channel check, matters more now than your gut ever will again. Up next: That Urgent Video From Your Boss Your Eyes Cant Catch The Fa.

---

The Part That Should Keep You Up at Night

Why People Search for Face Swap and Deepfake Tech Tutorials in the First Place

Here's the thing about the second-channel verification habit that nobody says out loud: by the time you verify that something was fake, the damage is often already done. A fake video of a politician goes viral at 8am. The correction runs at 4pm. Most people who saw the fake never see the correction. That's not a pessimistic take, it's just how information spreads.

Which means the only moment that actually matters is the moment before you hit share. Not after. Before. The verification habit has to happen upstream of amplification, not downstream of regret.

Piers Morgan, a journalist with decades of experience reading people, recently described at SXSW London how a deepfake of him fooled his own mother. His own mother. The woman who raised him couldn't tell. That's the level we're at. If a sharpened professional instinct and a lifetime of familiarity weren't enough protection for someone who knew him intimately, your ability to "just tell" isn't going to cut it either.

The question that should be sitting in the back of your mind right now isn't whether you could spot a deepfake of a politician you vaguely recognize. It's whether you'd catch one of someone you love.

People search for how to deepfake someone for a lot of reasons that have nothing to do with scamming anyone. Some want to understand deepfake software so they can recognize the seams when a fake lands in their own inbox. Others are curious how deepfake apps turned a task that used to require a film studio into something that fits on a phone. Whatever the reason, understanding the mechanics behind creating deepfakes is a reasonable way to demystify the threat instead of just fearing it.

At a basic level, most tools that make deepfakes rely on feeding a machine learning model a stack of photos or video of a real person's face. The model studies how that face moves, how light falls on it, how it forms expressions, and then learns to project those patterns onto footage of someone else. This is the same underlying idea behind every face swap tool, whether it is a novelty app on your phone or the more sophisticated deepfake tech used to build the convincing political ads described earlier in this piece.

The deepfake video pipeline usually starts with training data. The more real footage of a person's face and voice a model can study, from interviews, speeches, social posts, home movies, the more convincing the eventual deepfake becomes. That is exactly why public figures like politicians and celebrities are targeted so often. There is simply more raw footage of them online to train on, which is part of why deepfake methods keep improving fastest for the most famous faces first.

Voice cloning follows a similar logic to make deepfake audio convincing. As mentioned earlier, some tools now need only about five seconds of someone's real voice to generate a synthetic version that can carry a phone call. That is a much lower bar than most people assume, and it is exactly why the panicked phone call scenario in this article's headline is not hypothetical. It is already happening to families right now.

If you are trying to understand how to deepfake a video for research, journalism, or simple curiosity, it helps to know the tools generally fall into two camps. Consumer-grade deepfake apps handle simple face swap edits for entertainment, usually with visible watermarks or built-in limits. More capable deepfake software, the kind used to build the political ads and scam calls covered in this article, requires more technical setup but produces results convincing enough to fool a journalist's own mother, as Piers Morgan's story shows.

None of this technical detail changes the advice already given in this article. Knowing how deepfake video gets made does not make you faster at spotting one in the wild, because as covered above, even trained researchers and confident young adults were fooled at a rate of 99.9%. What understanding the process does give you is context, a sense of why a five second voice clip is enough, why political operatives target older voters, and why the tells that used to give away a fake are vanishing as the underlying deepfake technology improves.

So if your goal in learning how to deepfake is protective rather than curious, the verification habit already described in this article remains the actual answer. Understand the mechanics if it helps you sleep better, but do not mistake that understanding for a detection superpower. The second channel check, the callback, the independent search, the pause before you share, still does the actual work that your eyes and ears cannot.

Most people picturing an ai deepfake video imagine something that still needs a Hollywood budget, but the reality is closer to a laptop and an afternoon. That gap between public perception and actual accessibility is a big part of why the volume of deepfake video online keeps climbing every year. Every one of the videos described earlier in this piece, from the Kentucky ad to the New York attack clip, started as ordinary source footage before software reshaped it into something new.

It also helps to understand where images fit into the pipeline, since not every fake starts as a video at all. A single manipulated image, one photo passed through a face swap filter, can spread just as fast as a full clip and cause just as much confusion for someone scrolling quickly. Static images are often easier to produce convincingly than moving video, which is part of why a doctored image sometimes fools people who would catch a clumsier deepfake video.

The videos flagged in the UK polling data and the videos behind the Kentucky and New York ad campaigns did not require rare or specialized equipment. They used the same category of consumer-facing tools now marketed as deepfake apps, the kind you can find with a basic search. That accessibility is exactly why the scale of the problem grew so quickly in such a short window of time.

Understanding an image versus a full deepfake video matters because your verification habit should flex slightly depending on the format. A single suspicious image is easiest to check with a reverse image search, which can often surface the original unaltered photo in seconds. A deepfake video usually takes more digging, since you are looking for the original context, the original speaker, or the original event the footage was pulled from before it got altered.

Whether the fake in front of you is one still image or a full deepfake video, the underlying lesson from this entire article does not change. The technology producing both formats improves every year, the tells shrink every year, and the pause before you share is the one part of the equation that has not gotten harder. That is the part still fully within your control.

If you want to actually try deepfacelab faceswap style tools for yourself, purely to understand the pipeline described above, know that most of them still demand real patience. Setting up deepfacelab faceswap software typically means gathering hundreds of source images, training a model for hours or days on a decent graphics card, then manually cleaning up the output frame by frame. That barrier is dropping every year, but it is not yet as instant as a phone filter, which is exactly why the polished political deepfake video and scam calls covered earlier still tend to come from people who put in real setup time.

The faceswap concept itself is simple even when the software behind it gets complicated. A faceswap tool studies two faces, the source and the target, and learns to map the expressions, angles, and lighting of one onto the other so the result moves and speaks like the target person. Whether the output is a harmless novelty clip or a convincing deepfake used for fraud, the underlying faceswap math is the same, only the training time and image quality separate the toy version from the dangerous one.

Media literacy programs are starting to catch up to this reality, but slowly. Some schools and workplaces now include a short unit on how manipulated media spreads, covering both images and video, precisely because the old advice of just look closely stopped working. That shift in media education mirrors the shift described throughout this article, from trying to detect a fake with your eyes to building a verification habit that works regardless of how good the media gets.

Fraud built on this technology rarely announces itself as fraud. It arrives as an urgent voice message, a shocking video clip, or a photo that confirms something you already feared, which is exactly why the emotional hook matters more than the technical polish. Financial fraud teams increasingly describe deepfake based fraud as a social engineering problem wearing a technology costume, because the manipulation of your emotions is doing more work than the manipulation of the pixels.

Identity is the real target underneath most of this. A deepfake voice call is not really impersonating a sound, it is trying to borrow someone's identity long enough to extract money, information, or trust before anyone checks. Protecting your identity in this environment means treating any urgent request tied to a familiar voice or face, especially one demanding speed or secrecy, as unverified until you have confirmed it through a separate channel you control.

Facial recognition and identity verification systems are also adapting, though unevenly. Some workplaces now pair a video call with a live facial challenge, asking a person to turn their head or repeat a random phrase, specifically because a pre-recorded or generated face swap struggles to respond to unpredictable live instructions in real time. It is not a perfect fix, but it raises the cost of pulling off a convincing deepfake video scam, which is often enough to make a scammer move on to an easier target.

Images deserve one more note because they get less attention than video in most coverage of this topic. A single manipulated image shared in a group chat or comment section rarely gets the same scrutiny people apply to a full deepfake video, partly because a still image feels lower stakes and partly because reverse image search tools, while useful, are not something most people reach for automatically. Building the habit of checking a suspicious image the same way you would check a suspicious video closes a gap that a lot of current advice quietly skips.

Making a Deepfake Yourself Versus Recognizing One

There is a real difference between wanting to make a deepfake for a school project, a film experiment, or plain curiosity, and wanting to get better at recognizing one aimed at you. The mechanics covered in this article, the training data, the voice cloning, the faceswap math, apply to both goals equally. What changes is the point of the process you care about, either the creation stage or the moment someone else's creation lands in your inbox demanding a reaction.

If your interest is on the creation side, treat it as a technical hobby with real ethical lines, since using someone's face or voice without consent to make a deepfake can cause real harm even when the intent is a joke. If your interest is on the receiving end, everything this article has already said still applies, the pause, the second channel, the callback on a number you already trust. Either way, understanding how to deepfake something does not replace the verification habit, it just gives that habit better context to work from.