RN
Any of these platforms can produce a convincing picture now. That stopped being a differentiator some time ago, and it means the demo gallery on the front page tells you nothing useful — every service has a good one, because generating a good image is the easy part.
What actually separates them is a question the galleries never answer: who is the system willing to depict, and how does it know?
The line the category is really drawn along
There are two different products being sold under one banner, and conflating them is the single most common error in writing about this field.
Generation from description. You describe a person who does not exist and the system composes one. There is no subject, because no subject was ever involved. Whatever else is arguable here, consent is not — there is nobody to consent.
Generation from a likeness. The system starts from images of a specific real person and produces new material depicting them. Everything hinges on whether that person agreed, and in the overwhelming majority of real-world use, they did not.
The technical distance between the two is smaller than most people assume. Lightweight fine-tuning lets a general model be adapted to a specific face from a handful of photographs, at trivial cost, in minutes. A platform that offers character consistency has built most of the machinery required for the second use. Whether it is the second use comes down to what it permits, what it checks, and what it does when someone tries.
This directory treats that as a hard line rather than a spectrum. A service whose core function is producing sexual imagery of identifiable real people from ordinary photographs is not a variant of an AI porn generator; it is a different product with a different purpose, and it is not carried here.
The legal position, described honestly
It is moving, it differs by jurisdiction, and anyone who tells you it is settled is selling something.
The parts that are broadly stable: fully synthetic material depicting adults who do not exist is treated very differently from material depicting identifiable people, and a growing number of jurisdictions have legislated specifically on non-consensual synthetic intimate imagery — creating offences, requiring rapid platform takedown, or both. The direction of travel across the last few years has been consistently toward more restriction, not less, and the obligations increasingly land on platforms rather than only on individual users.
The parts that are genuinely unsettled: liability for training on material without a licence, the status of likenesses that resemble a real person without being derived from them, and how ordinary jurisdictional conflicts resolve when the user, the platform and the depicted person are in three different countries.
None of that is legal advice, and if you have a specific situation, none of it substitutes for someone qualified to look at it.
What to actually check before signing up
Skip the samples. The following take a few minutes and separate serious operations from the rest.
Is there a stated likeness policy at all? Not a general terms-of-service paragraph about "illegal content" — a specific statement about real people, celebrities and uploaded photographs. Its absence on a platform that accepts image uploads is the loudest signal available.
Does it accept uploaded reference images, and under what conditions? Upload capability is not disqualifying — it is how you produce a consistent original character. But a platform that accepts arbitrary photographs of faces with no gate on whose face it is has made a choice, and that choice is the whole product for some of them.
Is there a takedown route for a depicted person? Note the distinction: a route for someone who is not a customer to report material depicting them. A support address for account holders is not the same mechanism and does not serve the same people.
What does it say about training data? Very few publish specifics and you should not expect a dataset manifest. But there is a meaningful difference between a platform that addresses provenance and one that has never mentioned the subject.
Is generated output marked? Visible watermarks, embedded metadata, or the C2PA provenance standard that records how a piece of media was made and travels with the file. Marking is imperfect — metadata can be stripped, watermarks cropped — and it is still the difference between a platform that considers downstream misuse its problem and one that does not.
What happens to your prompts? A prompt is a more explicit statement of what you want than any browsing history, arriving as plain text. Retention, human review and training reuse are all worth knowing about before you type the first one.
Where the ordinary consumer questions still apply
Once the above is satisfied, the boring commercial checks matter as much here as anywhere:
- Credits, not subscriptions. Most of these run on generation credits. Find out what one credit buys, whether unused credits expire, and what a failed generation costs — some charge for outputs you discard.
- Rights in the output. Whether you may use what you generate, and whether the platform claims a licence over it, varies enormously and is usually buried.
- Does anything you make persist? Some galleries are private, some are public by default, and some are public by default with an opt-out you have to find.
- Cancellation. The recurring complaint across the whole adult subscription economy, generators included, is billing that continues after a cancellation the user believed had gone through.
The honest summary
The technology is impressive and the interesting uses are real — specificity you cannot find in any catalogue, iteration instead of search, aesthetics nobody was funding. None of that is in dispute.
But this is the one adult category where the main risk is not to the buyer. Everywhere else, a bad site wastes your money. Here, the same machinery that makes a good product makes a weapon, and the only thing standing between them is what the operator decided to permit. That makes the policy page the review, and the gallery decoration.
If you are shopping the category, the generator listings and the wider AI section are the starting point — and the entries there are researched against these questions, not against how good the samples look.