Other People's Faces in Generation: Where the Line Runs

The ability to create a plausible image of a person who was never in that situation is the most problematic property of generative technology. The restrictions around it exist not from abstract caution but because of concrete harm that has already occurred.

What the real harm is

• Intimate imagery without consent — the most widespread abuse, directed predominantly at women.
• Fraud: fake video messages purporting to come from executives or relatives, used to obtain money.
• Disinformation: images of events that never happened, in political and public contexts.
• Harassment: compromising images of specific people, created for distribution.
• Forged evidence: images used as proof.

What all these share is that the image works because it looks like a photograph, and photographs are traditionally received as evidence.

Why resemblance creates the problem

The formal argument "this is not a photograph of that person but a generated image" fails for a simple reason: the harm comes from recognisability, not from the file's origin.

If a viewer identifies the person, the image acts exactly as a real photograph would: it forms opinions, affects attitudes, spreads. The technical nature of the file is irrelevant to that.

What services restrict

Practically every generative service closes off the creation of images of real people in invented situations, especially public figures. The reason is scale: mass availability turns isolated abuse into a systemic problem.

Restrictions usually extend to uploading other people's photographs without consent — the first barrier and the easiest to enforce.

What responsible use looks like

1. Upload only your own photographs, or others' with explicit permission.
2. Do not create images of recognisable people in situations that never happened.
3. Do not publish results in ways that could be taken for documentary photography.
4. Label generated material where it could mislead.
5. Do not attempt to circumvent a service's restrictions.

Consent

The key element, often reduced to a formality. Consent means the person knows what exactly will be created and where it will be used.

"He doesn't mind me uploading his photo" and "he agrees to his image appearing in an advertisement" are different things. The second requires an explicit conversation, especially where the result is published.

What to do if you are the target

If you discover a generated image of yourself made without consent, the standard steps are:

1. Document it: save links, screenshots, the date of discovery.
2. Report it on the platform hosting the image: most have a mechanism for non-consensual imagery, and it usually works quickly.
3. Contact the service if you know where it was created.
4. For serious harm, contact law enforcement: in many jurisdictions such acts are punishable.

The technological side

Mechanisms for confirming image provenance are developing: cryptographic signatures, embedded marks, systems tracking a file's history. They do not solve the problem entirely, but they raise the cost of abuse and make investigation easier.

Detection tools are improving in parallel. Neither direction offers complete protection, but together they shift the balance.

Why this concerns ordinary users

The topic seems to belong to bad actors, but in practice the line runs closer. A joke image of a colleague, a greeting collage with someone else's face, "trying on" a look for a friend — all of it is creating an image of a person without their involvement.

A practical rule: before uploading someone else's face, ask yourself whether you would show them the result. If yes, it is almost certainly fine. If no, that is your answer.