RN
Most arguments about AI porn stall on the wrong question. Someone asks whether it is real, someone else answers that of course it is not, and both walk away having settled nothing — because "is it real" is not the question anyone was actually worried about.
Sort the field by who gets hurt and it separates cleanly into three problems. They have different victims, different legal treatment, and different solutions, and conflating them is why the discussion goes nowhere.
Problem one: material about people who did not agree
This is the serious one and it has nothing to do with whether synthetic imagery should exist.
Generative tools made it cheap and fast to produce sexual imagery of an identifiable person from ordinary photographs of them. Not a celebrity with a large public image corpus — anyone, from a handful of social media pictures. The barrier used to be skill and time; there is no barrier now.
The victims of this are overwhelmingly not public figures. They are classmates, colleagues, ex-partners and strangers, and the harm is not hypothetical or reputational-in-theory. It is the ordinary catastrophic harm of intimate imagery circulating without consent, with the additional cruelty that the target cannot even say it did not happen in a way that helps, because the image exists regardless of what it depicts.
The law has moved on this faster than on almost anything else in the space. Jurisdictions that already criminalised sharing intimate images without consent have extended those offences to synthetic material, and several have created specific offences for producing or distributing sexualised deepfakes of identifiable people. The direction is uniform even where the detail is not.
The tools built specifically for this purpose — the ones whose function is taking a photograph of a person and generating a nude version of them — do not have a legitimate consent-respecting use case that their design supports. Consent is not a policy those services fail to implement. It is structurally absent from the product.
"Nobody was harmed because nobody was real" does not apply here. Someone real is the subject. That is the whole point of the output.
Problem two: material depicting people who cannot consent
Generated material depicting minors is the second problem and there is nothing to discuss about it. A growing number of jurisdictions have explicitly extended child sexual abuse material offences to synthetic depictions, on the correct reasoning that the harm — normalisation, use in grooming, the market it sustains — does not require a specific child to have been filmed.
Two adjacent facts matter. Models trained on scraped web data have in documented cases ingested illegal material, which is a supply-chain problem for anyone building or deploying them. And services that generate this on request are prosecutable as producers, not merely as hosts.
There is no version of the "it's just pixels" argument that survives contact with this category, and anyone advancing it in the first problem should notice it is the same argument.
Problem three: everything else
Which leaves the large remainder: entirely synthetic material depicting nobody in particular, made by people who wanted to make it, consumed by people who wanted to see it.
This is where the actual debate is, and it is a much smaller and more ordinary debate than the first two suggest. The concerns are real but they are not emergencies:
Displacement. Synthetic content competes with work by performers who need the income. That is a labour question of the kind every creative industry is currently having, and it deserves the same seriousness and no more panic.
Likeness leakage. A model trained on real performers can produce outputs that resemble them without naming them. The person's face is doing commercial work and the person is not being paid. This sits between problems one and three and is genuinely unresolved.
Disclosure. An audience that cannot tell what it is looking at is being misled, which is a consumer-honesty issue rather than a safety one.
Calibration. Synthetic material optimises for what engages, without the constraints of physics, fatigue or a person's limits. What that does to expectations over time is an open question and honest people should say it is open rather than asserting an answer.
None of these justify treating the whole category as an emergency. All of them are worth arguing about properly.
What "getting it right" would look like
For services: verifiable consent for any likeness of a real person, refusal of identifiable-person prompts by default rather than on complaint, provenance metadata attached at generation, and a takedown route that a distressed non-technical person can actually use. Most of these are engineering decisions, not impossible ones — a service that has not made them has chosen not to.
For creators: disclose synthetic material as synthetic. Do not build likenesses of real people without their agreement, including the "inspired by" kind that everyone recognises. And keep records of what you generated and how.
For viewers, the honest advice is narrower than the confident advice you will see elsewhere. Detection tools exist and none of them are reliable enough to stake anything on — the visual tells that worked eighteen months ago are largely gone, and any list of them dates immediately. What holds up is provenance rather than forensics: whether the material states what it is, whether the platform has a consent policy, and whether an identifiable real person appears to be depicted.
That last one is the only test that matters most of the time. If it looks like a specific real person and there is no indication they agreed, that is the line, and it does not require you to identify a single rendering artefact.
Where the line actually is
Not at "synthetic". At consent — of the person depicted, and of the people whose work and likeness went into making the depiction possible.
Everything upstream of that line is a normal argument about a new medium, and it will resolve the way those arguments usually do. Everything downstream of it is a harm being done to somebody, and calling it a fantasy does not change who is on the receiving end.