Are AI porn generators uncensored? What each one bans

Almost every AI porn generator advertises itself as uncensored or "no filter." Read the fine print and every one of them has a list of things it will refuse to make, and a system that watches what you type and what the model produces. The list is short and nearly identical across the industry: nobody who is or looks under 18, no real people, no non-consent, no animals. Where the sites differ is in how strictly they define those lines and how they enforce them. So no AI porn generator is truly uncensored, and the ones that claim to be are describing their marketing, not their terms.

This guide quotes the companies' own policy pages, which we loaded on November 10, 2026, explains how the filters actually work, and walks through the cases where moderation in this industry has failed, sometimes badly. Explicit content between fictional adults is legal for adults in most places. The things these sites ban are banned because they are illegal or because they hurt real people.

What do AI porn generators ban?

All seven we review ban anything involving minors or people who look under 18, real people without consent, and non-consensual scenes; most also ban animals and uploads of other people's photos.

Minors, and anyone who looks like one

This is the one rule with no exceptions anywhere, and the better policies define it by appearance, not by a stated age. Promptchan's content policy bans "any depiction of a person under 18, or of anyone who appears to be under 18, in any context, including a non-sexual one," and names youthful faces or bodies, school settings, role-play and illustrated characters with childlike features. OurDream's prohibited content page says "We judge age on the whole picture, not a stated number," and lists "about to turn 18," childlike settings and loli-style art "whatever the stated age."

Secrets.ai's Prohibited Content Policy (updated April 10, 2026) specifically bans "attempts to age-down characters to circumvent restrictions" and adds a line worth remembering: trying to generate prohibited content, "even if unsuccessful," counts as a violation in itself. Candy.ai's Blocked Content Policy says it is "strictly prohibited trying to generate AI content resembling minors," and Xotic AI's moderation guide rules out "'aged-up' child characters." This is also the law: sexual imagery of minors is a crime whether a camera or a model made it, as our guide to AI porn laws explains.

Real people

Every site bans sexual images of identifiable real people, but the wording varies. Promptchan is the strictest: no depiction of a real, identifiable person, public or private, "even with their consent." OurDream bans characters "recognisably modelled on a real person (alive, or deceased less than 100 years)," including fictional characters whose appearance "closely matches a specific real actor or model." Secrets bans sexual content of real people "without their verified consent." Candy has "a zero-tolerance policy for the creation or attempted creation of non-consensual intimate imagery and images intended to resemble actual persons ('deepfakes')." Xotic's Prohibited Content Policy bans "AI-generated depictions of real people, including celebrities, without consent."

DarLink's Terms and Conditions (last updated August 2025) tell users not to impersonate "a celebrity or public figure," but in the version we read we did not find a clause aimed specifically at generating images of real people.

Uploading someone else's photo

The fastest way to make a deepfake is to upload a real photo and ask the tool to undress it, which is exactly what "nudify" apps do. Most of the platforms we review close that door. OurDream's help center says: "We do not allow for the uploading of reference photos to OurDream due to possible deepfakes and other legal problems." Candy says users "cannot upload or share content on the Services." Secrets' Content Removal Policy says "No user-uploaded images or real-person media are used."

Lovescape's Content Moderation Policy takes a middle path. Ordinary users cannot upload photos, videos or audio to chat with characters. Pro subscribers can upload, but only after identity verification and a checkbox confirming they own the rights, and every upload gets human review within 24 hours before it can be shared. A likeness complaint triggers "immediate takedown and blocking."

Non-consent, animals and the rest

Promptchan bans sexual content that "depicts or suggests a lack of consent," including someone asleep, unconscious, intoxicated or incapacitated, along with incest, animals and death in a sexual context. Secrets bans non-consensual content even in "scenarios framed as 'fiction.'" DarLink bans "bestiality, rape or non-consensual sexual violence." Xotic lists zoophilia, necrophilia, incest and cannibalism. OurDream draws a finer line in text role-play: a user may play the receiving side of a dubious-consent fantasy, but not a framing that casts the user as the perpetrator or presents the lack of consent as the appeal.

So "uncensored" in the marketing means nudity and explicit sex between adults are allowed. It never means these categories are.

How do AI porn filters work?

In layers: a filter reads your prompt before anything is made, a classifier checks the image after, and flagged accounts go to human reviewers.

Before anything is generated: prompt filters

The first check runs on your words. Promptchan says "an AI moderator and a list of banned words check every prompt and chat message," and a request that breaks the rules is refused with a reason. Secrets' Safety Guidelines describe "contextual pattern detection to identify indirect or coded attempts to elicit prohibited content." Xotic's Content Moderation Policy says the system answers with "a safe pre-prepared message and disengages." Lovescape says that if someone tries to build a character based on a real individual, "the AI modifies the request to prevent unauthorized imitations."

Everyone knows people try to sneak past these lists. OurDream names the tricks it looks for, including non-English prompts, base64 and other encoding, tags designed to evade detection and "repeated near duplicate submissions," and treats evasion as the underlying violation plus a separate breach of its rules.

Word filters are blunt, so they also catch innocent messages. Candy's help center admits that "messages may sometimes be flagged by mistake." OurDream is more candid about why: "Our system is a bit too sensitive right now and may flag safe content," and "We have to keep the system strict to comply with our third-party payment providers." Card networks and payment processors are a big part of why these filters exist at all.

After generation: image classifiers

A clean prompt can still produce a bad image, so the better systems check the output too. OurDream's moderation overview describes a separate set of classifiers that look at each generated image "independent of the prompt." Promptchan says its AI moderator checks every generated picture, "including for age," and a picture that fails "isn't delivered, saved or published." Secrets says all outputs, including text, images and video, go through automated classifiers before delivery, and anything that trips them is withheld.

Name blocklists and face matching

Real-person filtering is harder than it sounds. A name blocklist catches a celebrity's name typed into a prompt, but not a careful description of her face. OurDream says it keeps a "real-person blocklist" of names, and lists matches against it among the four most common reasons it rejects public submissions. Face matching, which compares a face in an upload or an output against images of known people, is the stronger tool. None of the seven sites we review publishes how, or whether, it matches generated faces against real people, so we can't tell you how well any of them does it.

Behavior signals and human reviewers

Automated checks see one request at a time. OurDream adds a layer that watches patterns across many actions, such as accounts whose requests are rejected at unusually high rates, which raises their priority for human review. Lovescape escalates a user to a human moderator automatically after three or more attempts to break the rules. Promptchan says "a person decides every suspension and ban," apart from card chargebacks, and keeps records of refused requests for 90 days in most cases.

On OurDream, the most common rejection in the public-submission queue is "profile images with underage cues," ahead of copyrighted characters, non-consent framings and real-person names. That ordering is a useful reminder of where the pressure on these systems actually is.

Is a private chat really private?

Not from the machines. OurDream says its classifiers process all content on the platform, "including private chat messages and private generations," while human reviewers see only content that was flagged, reported or part of an investigation. Candy says conversations are "generally confidential" but run through moderation built on its own language model, and that when something is flagged staff may review it "and/or other content associated with the account." Our AI porn generator privacy guide covers who else can see your prompts, and what has leaked.

Do AI porn generators verify your age?

Mostly with a click. Real ID or selfie checks appear only where the law requires them, such as the UK and Australia.

For most sites the age check is a click. Lovescape asks users to "self-declare" that they are at least 18 during registration. Xotic's Underage Policy describes an age gate where users confirm they are 18 or older. DarLink says it may suspend an account "pending age verification" if it has reason to believe the user is under 18.

Real verification shows up where the law demands it. OurDream's age verification page says users in the United Kingdom and Australia, among others, must verify through a third-party provider, Go.cam or AgeVerif, using methods that can include AI age estimation from a selfie, a government ID or a credit card. OurDream says it receives only a pass or fail signal and never the ID or the selfie. Promptchan puts every new account in Safe Mode, shows only the safe version in some countries, and asks for proof of age "where the law or we ask for it." Secrets takes a different route: its Terms of Service say the site is meant for a territory that "excludes all jurisdictions where the content is illegal or subject to age verification measures."

Reporting, appeals and what happens to accounts

Reporting content

Promptchan puts a Report button on every published picture and video and runs a Report Content form that anyone can use, with or without an account. Lovescape promises a confirmation email and an outcome within seven business days. Secrets reviews reports within five business days and escalates anything involving possible child abuse material immediately. Its real-person removal process may ask for proof of identity and aims to finish within five business days of a complete request. DarLink says its team typically reviews reports within 48 hours.

If someone has made a fake sexual image of you, on any site, US law now gives you a faster route. The TAKE IT DOWN Act, signed May 19, 2025, required covered platforms to set up a notice-and-removal process within a year and to remove reported intimate images within 48 hours of a valid request. Our takedown guide walks through the steps.

What happens when you break the rules

For minor-related violations the answer is the same everywhere: the account is closed and the attempt can be reported. Secrets' protocol is the most detailed: the output is blocked, the account flagged for immediate human review, terminated on confirmation, and the content and account information reported to the National Center for Missing and Exploited Children (NCMEC) with evidence preserved. Candy says it reports known CSAM to NCMEC. OurDream names NCMEC's CyberTipline, the UK's Internet Watch Foundation and Australia's child exploitation center. Promptchan says suspected child abuse material, including AI-generated material, "is preserved securely as evidence, with the related account records, rather than deleted."

Lesser violations get graduated treatment. Xotic gives a strike for each violation on shared surfaces like its feed, and "three strikes within 90 days" mutes the account for seven days. Promptchan moves from refusal to warning to human review to suspension or ban. DarLink says an account terminated for a violation gets no refund.

Appeals

Promptchan gives users six months to appeal by email; someone who wasn't part of the first decision looks again, usually within seven days. OurDream sends appeals to a different reviewer. Lovescape is the outlier: its policy says "all moderation decisions are final."

When moderation fails

The training data itself

In December 2023 the Stanford Internet Observatory published Identifying and Eliminating CSAM in Generative ML Training Data and Models, an audit of LAION-5B, a dataset of billions of web images used in part to train Stable Diffusion. Using hash matching and other methods, the researchers identified 3,226 dataset entries of suspected CSAM, "much of which was confirmed as CSAM by third parties." Filters on top of a model cannot fully fix what went into it.

A mainstream chatbot that undressed people

In late December 2025 and early January 2026, users of xAI's Grok on X used its image editing to put real women and children into revealing clothing without their consent. California Attorney General Rob Bonta opened an investigation on January 14, 2026, citing an analysis that found more than half of roughly 20,000 images Grok generated between Christmas and New Year's showed people in minimal clothing, some of whom appeared to be children. According to the Associated Press, xAI then said it had added "technological measures" to stop the Grok account from editing images of real people into revealing clothing, but AP found the editing tool still offered to free accounts the next morning, and Malaysia and Indonesia had blocked Grok.

The biggest deepfake porn site closed

Mr. Deepfakes, the largest site for sexual deepfakes of real people, went offline in May 2025. Its notice, reported by 404 Media, said "A critical service provider has terminated service permanently" and "We will not be relaunching." Which provider, and why, was not disclosed.

Nudify sites in court

San Francisco City Attorney David Chiu sued the operators of 16 of the most-visited "undressing" websites in August 2024. By June 2025, 10 of them were offline or blocked in California, and one operator, Briver LLC, had agreed to a permanent injunction and $100,000 in civil penalties. The case continues against the remaining defendants.

Meta sued Joy Timeline HK, the company behind the CrushAI nudify apps, in Hong Kong in June 2025, after what Meta called repeated attempts to get around its ad review. Meta said it had shared more than 3,800 nudify-related URLs with other tech companies since late March 2025.

The scale of the problem

NCMEC reported a 1,325% increase in CyberTipline reports involving generative AI in 2024. Part of that is more companies reporting, which is the system working. The rest is why the rules above exist.

What this means for you

• Don't try to get around the filters. On several of these sites an attempt is a violation on its own, it is logged, and attempts involving minors can be reported to authorities.
• Never upload or describe a real person. Most of these platforms don't take uploads at all, and the ones that do verify who you are first.
• If an innocent prompt gets flagged, use the report or appeal route. False positives are common and the sites say so.
• Assume your private generations are scanned by software, and that a human may read them if they're flagged.
• If you find content that involves a minor or a real person without consent, report it to the site and, for minors, to NCMEC's CyberTipline.