An ai video of real people sits in a gap that most rules have not caught up with. Generating a clip from your own photo is unremarkable. Generating one from someone else's photo without asking is a different act entirely, even when the file was public and even when the intent was harmless.
The framework below sorts four scenarios, because "is this okay" has four different answers depending on who is in the frame.
The four scenarios
| Who is in the frame | Who uploaded the photo | Verdict |
|---|---|---|
| You and your partner | You | Fine |
| You and your partner | Someone else, with permission | Fine |
| A friend or ex-partner | You, without asking | Not fine |
| A public figure | Anyone | Depends on purpose, usually not fine |
The variable that matters is not legality but whether the person depicted would recognise the act as one they agreed to. A clip of a friend, made privately and never published, is a different thing from the same clip posted publicly — but both require that the friend knows.
Consent is not a one-time permission. It is an ongoing state that can be withdrawn.
That single line resolves most of the hard cases. If someone asks you to take a clip down, the correct response is to take it down, not to explain why the request is unreasonable.
Why an ai video of real people is not only an etiquette question
Deepfake regulation is moving quickly and unevenly. The EU's AI Act introduces transparency obligations for synthetic content, including requirements to disclose that media has been artificially generated. UNESCO's Recommendation on the Ethics of Artificial Intelligence sets out the wider principle that has been adopted by a large number of member states: proportionality, do-no-harm, and human oversight.
In practice, three rules cover almost everything a normal person will do:
- Disclose. If the clip could be mistaken for real footage, label it.
- Do not impersonate. Never present generated footage as a real recording of a real event.
- Do not depict harm to a named person. This is the line that turns a prank into an offence in many jurisdictions.
The NIST AI Risk Management Framework is written for organisations rather than couples, but its core move is useful at any scale: identify who could be harmed before you ship, not after.
The specific case everyone asks about
Using a photo of an ex-partner. No. Even privately, even as a joke, even if the photo is still on your phone. This is the single most common version of this problem and the answer does not vary by jurisdiction.
Using a photo of a deceased relative. Ask the immediate family first. Some will find it comforting; others will find it a violation. There is no default.
Using a photo of a public figure for satire. Satire has protections in many places, but synthetic footage of a real person saying something they never said is not automatically satire. The disclosure rule applies.
Using a stock photo of a stranger. Check the licence. Most stock licences permit editorial use, not the creation of new synthetic media featuring the model.
A five-question check before you generate
- Did the person in this photo agree to this use, not just to being photographed?
- If this clip were published without context, could someone think it is real?
- Would I be comfortable showing it to them in person?
- Does the platform I am posting to require an AI label?
- Do I have the source photo and the settings, in case I need to prove how it was made?
Any "no" in the first three is a stop. The last two are housekeeping, and cheap: a render costs 8 credits, and the $7.99 entry pack covers the attempt. Nothing about the cost justifies skipping the question.
What responsible use looks like in practice
It looks boring. You make clips of people who asked for them, you label them when they could be mistaken for real footage, and you delete them when someone asks. Disclosure costs nothing: a clip is 4 to 6 seconds long, and the label is one toggle.
The interesting part is what this enables. Because the technology is cheap and fast, the constraint on how much you can make is entirely social rather than technical — and that is a good thing. An ai video of real people is only worth making when the people in it are glad it exists.
Frequently asked questions
Is an ai video of real people legal if I took the photo?
Taking the photo gives you rights over the image; it does not automatically give you the right to generate new synthetic media from it, particularly if you publish it. Consent from the person depicted is the safer basis in every jurisdiction.
Do I need to label a clip made from my own photo?
If it could be mistaken for real footage, yes — most platforms now provide an AI-generated toggle. A stylised zombie clip is unlikely to be confused with a recording; a realistic one might be.
What about photos of my own children?
Parental consent covers your own children, but publishing is a separate decision. Many platforms restrict synthetic media depicting minors, and the safe default is not to publish.
Can someone make a clip from my public photo?
Technically yes, which is why the norms matter more than the tools. Most jurisdictions now provide a route to have such content removed, and platforms act on reports of non-consensual synthetic media.
If everyone in the frame asked for the clip, you are in the simple case, and it is most cases. Make it — and if anyone in the photo has not seen it yet, show them before you post.