
Contents
- 1. How to read a case
- 2. The viral celebrity video
- 3. The school incident
- 4. The election robocall
- 5. The cloned executive voice
- 6. The fake endorsement ad
- 7. What it costs to do it right
- 8. Honest limitations
- 9. הסכמה: the line that does not move
- 10. FAQ
1. How to read a case
A case is not a story to retell for shock value. It is a record of what happened when a rule was broken, and the rule is usually simple. This guide treats five widely reported patterns from the last few years. Names, exact dates, and quotes are left out on purpose — the details that matter are the pattern and the consequence, and the people harmed in these cases have already had enough attention. For each case, ask the same three questions: whose face or voice was used, did they agree, and what happened once the file left the creator’s computer? If any answer is uncomfortable, that discomfort is the lesson. In every case below, the case is the lesson, and the lesson is the same: permission first, or do not make it.
One more note on how this guide is written. Nothing here is legal advice. Laws differ by country and change fast — the round-3 guide on this site covers the law by country in detail. What follows is about judgment: what creators can learn from other people’s mistakes before making their own.
| Case pattern | What was used | Who was harmed | The lesson |
|---|---|---|---|
| Viral celebrity video | A famous person’s face, explicit content | The person depicted; every viewer misled | Virality outruns takedown |
| School incident | Classmates’ photos, fake explicit images | Minors at school; the whole community | Non-consent plus minors is criminal exposure |
| Election robocall | A candidate’s cloned voice | Voters; trust in the vote | A voice is not proof of identity |
| Cloned executive voice | A CEO’s voice on a call | An employee; the company | Verify through a second channel |
| Fake endorsement ad | A public figure’s likeness in an ad | The figure; buyers who trusted it | Likeness is not free stock |
2. The viral celebrity video
Early in 2024, explicit videos face-swapped onto a world-famous singer spread across social platforms in a matter of hours. The images were fake, the singer had nothing to do with them, and the platforms took days to catch up — by which time the files had been copied, reposted, and discussed everywhere. The coverage that followed was about the singer, the platforms, and the law, but the part creators should remember is the speed. A file made in an evening can travel further in a day than its maker ever imagined, and once it is out, no delete button reaches every copy.
The lesson is not “do not get caught.” The lesson is that distribution is the harm. The person in the video did not consent, and every share — even a share that says “look how fake this is” — spreads the image further and tells the next viewer that this is a thing people do. The case is the lesson: if the face is real and the consent is not, the file should never exist. A creator who wants to practice explicit face-swap technique has a legitimate alternative — a fully fictional adult character, built from scratch, whose face belongs to no one. The technique is the same. The harm is gone.
This case also changed the law in several places. The United States’ TAKE IT DOWN Act (2025) requires platforms to remove non-consensual intimate imagery quickly when notified. The UK and the EU moved in the same direction. None of that helps the person in the video after the fact. It only punishes the maker after the damage is done.
3. The school incident
In several widely reported incidents, students used AI tools to generate fake explicit images of their classmates — often from ordinary school photos — and shared them in group chats. The images were fake. The harm was not. The students depicted faced humiliation at school, the schools faced police involvement, and the students who made the images faced expulsion and criminal investigation in some jurisdictions.
Three things make this pattern different from the celebrity case, and worse. ראשון, the victims are minors. Non-consensual intimate imagery of a minor is treated as child sexual abuse material in most legal systems, regardless of whether the image is “real” or AI-generated — the fact that no real body was photographed does not make it legal. שני, the victims cannot escape the audience: the same people who see the images sit next to them in class the next day. שלישי, the makers are often minors themselves, which means a stupid decision at fifteen can become a criminal record that follows them for years.
The lesson for creators is blunt: never generate explicit imagery of anyone under 18, never generate explicit imagery of a real person without consent, and never assume that “it was just a joke between friends” is a defense. It is not. If you make adult content, make it about adults — fictional adults you created, or yourself.
4. The election robocall
Before a primary election in the United States in 2024, voters in one state received robocalls in the voice of a leading candidate, telling them not to vote. The voice was a clone. The candidate had said no such thing. The calls were widely reported, investigated, and eventually traced, but the damage was done in the hours before the vote, when the message was still circulating and no correction had caught up with it.
The lesson for creators is about a different kind of consent: a person’s voice is their identity, and cloning it to put words in their mouth is a forgery, even when the words are not explicit. The technology that makes a good voice clone is the same technology that makes a good parody — the difference is not the tool, it is the target and the purpose. A fictional character with a fictional voice is a creation. A real candidate’s voice telling voters to stay home is election interference.
There is a practical lesson here too, for everyone who receives media rather than makes it: a voice on a call is not proof of who is speaking. If a voice tells you to do something consequential — vote, pay, transfer, share a password — verify through a second channel before you act. The case is the lesson, and the lesson applies to audiences as much as to creators.
5. The cloned executive voice
In a widely reported fraud case, an employee at a company received a phone call from someone who sounded exactly like the chief executive, instructing them to transfer a large sum of money. The voice was a clone built from public recordings. The employee followed the instruction. The money went to the fraudsters.
The technical lesson is that a few seconds of public audio is enough to build a convincing clone, and a phone call gives the listener almost no way to check. The human lesson is that authority plus urgency is a powerful combination: when the boss’s voice says “do it now,” most people do it now. The defense is procedural, not technical — no large transfer on a single voice instruction, always a second channel, always a callback to a known number.
For creators, the point is narrower: do not clone real people’s voices, even for a joke, a test, or a demo. A voice clone of a real person is a forgery waiting for a use, and the use is rarely yours to control once the file exists. If a project needs a voice, use your own, hire a voice actor, or use a synthetic voice that belongs to no one.
6. The fake endorsement ad
Another recurring pattern: an ad appears on social media showing a well-known public figure apparently endorsing a product — a supplement, a crypto scheme, a weight-loss plan. The figure never endorsed it. The face is a swap or a generated image, the quote is invented, and the ad is designed to borrow trust the product has not earned. Buyers who trusted the face lost money. The figure spent time and lawyers getting the ads taken down.
The lesson is that likeness is not free stock photography. A famous face is not a prop you can borrow because it is public. Endorsement law in most countries treats a fake endorsement as deception, and the platforms that run the ad are increasingly liable for it. For creators, the safe version of this idea is obvious: invent a fictional spokesperson, or be the spokesperson yourself. A fictional character endorsing your own product is honest. A real person’s face endorsing a product they never touched is fraud.

7. What it costs to do it right
The site itself offers a free image preview per IP with no sign-up, so you can test the generator before spending anything. Doing this ethically is not expensive. Checking a suspicious image costs nothing — free detection checkers, reverse image search, and metadata readers are all free, and the round-4 guide on this site walks through all three. Rebuilding a scene the right way, with a fictional face instead of a real one, costs one face swap: 26 קרדיטים, which is $0.52 $19.99 חבילה ו $0.31 $59.99 חבילה. Three attempts cost 78 קרדיטים: $1.56 או $0.94. An image-to-image pass to fix a broken area costs 49 קרדיטים: $0.98 או $0.59.
Compare that with the cost of doing it wrong: a takedown, a banned account, a legal letter, a criminal investigation, or a reputation that follows you for years. The ethical version is cheaper in every currency that matters. The case is the lesson, and the cheapest lesson is someone else’s.

8. Honest limitations
This guide generalizes from public reporting, and generalization has limits. The cases are described as patterns because the details — names, dates, exact quotes — are deliberately left out; that means you cannot verify a specific claim here against a specific news article. The law summaries are summaries, לא ייעוץ משפטי, and the law is moving fast: what is true in one country may be false in another, and what is true this year may change next year. If you need a legal answer for a specific situation, talk to a lawyer in your jurisdiction.
Detection is also limited. No tool can prove an image is real or fake with certainty — compression destroys evidence, new generators beat old detectors, and real photos can fail checks that fakes pass. The round-2 guide on this site covers the seven signals the human eye can check, and the round-4 guide covers the tools. Use both, and remember that uncertainty is a reason to stop, not a reason to publish.
9. הסכמה: the line that does not move
Every case in this guide broke the same rule, so here it is once more, without exceptions: only generate intimate or explicit imagery of yourself, or of fully fictional adult characters you created. Never use a real person’s face, voice, or likeness without their explicit consent — not a celebrity, not a classmate, not a colleague, not a politician, not a stranger whose photo you found online. “They are famous” is not consent. “It is just a test” is not consent. “Nobody will know” is not consent.
Consent is not a formality. It is the difference between creation and forgery, between art and harm. The cases above are what happens when creators decide the rule does not apply to them. It applies to everyone, every time, including you.
10. FAQ
Is it illegal to make a deepfake of a real person?
It depends on the country and the content, but non-consensual intimate deepfakes are illegal in many places, including the United States under the TAKE IT DOWN Act (2025), the UK, and the EU. This guide is not legal advice — check the law in your jurisdiction.
What if the image is obviously fake — does that make it okay?
לא. Obviousness does not remove the harm: the image still spreads, still misleads some viewers, and still uses a real person’s likeness without permission. A fictional face avoids the problem entirely.
Can I use a celebrity’s face for a parody?
Parody law is narrow and varies by country, and it does not cover explicit or sexual content. Do not use a real person’s face for explicit imagery under any label — parody, satire, or art.
How do I practice face swap ethically?
Use your own face, or a fully fictional adult character you created. One face swap costs 26 קרדיטים ($0.52 $19.99 חבילה, $0.31 $59.99 חבילה). מחירים נבדקו 13 נוב׳ 2026.
What should I do if I see a non-consensual deepfake?
Do not share it — sharing spreads the harm even with a warning. Report it to the platform, and if the person depicted is identifiable, use the platform’s non-consensual imagery reporting channel.
Do detection tools prove an image is fake?
לא. No tool proves real or fake with certainty. בודקים חינמיים, חיפוש הפוך, and metadata readers give hints, not verdicts — the round-4 guide on this site explains all three and their limits.