AI porn is sexually explicit images, video, or audio generated by AI, spanning consensual synthetic content and non-consensual deepfakes targeted by 2026 law.
Last verified · 2026-10-23 · by Moe Ameen
Published 2026-09-11 · Updated
Short answer
AI porn is sexually explicit images, video, audio or text made or heavily altered by generative AI instead of filmed. Consent splits it in two: disclosed content a performer authorized, and non-consensual deepfakes of real people. A 2019 Deeptrace report found 96% of 14,678 deepfake videos online were pornographic. The US TAKE IT DOWN Act, signed May 19, 2025, criminalizes publishing non-consensual intimate images.
Key takeaways
Porn AI — more commonly written "AI porn" — is a catch-all term for sexually explicit media produced or heavily altered by generative AI rather than filmed with real performers. It covers still images from text-to-image models, short explicit video clips, cloned or synthesized voices, and text-based erotica from chatbots. The defining feature is that no camera and, in the fully synthetic case, no real body was involved: the content is generated from a prompt, a reference image, or both.
The term spans two very different things that are worth keeping separate. On one side is consensual, disclosed synthetic adult content — fully fictional characters, or media a real performer authorized and often generated of themselves to scale their own output. On the other is non-consensual material: sexual images or video of a real person, made without their permission by feeding a photo of their face into a generator. That second category is where the term earns its notoriety. It includes celebrity deepfakes, "nudify" apps that strip clothing from an uploaded photo, and — the most serious and universally illegal case — AI-generated child sexual abuse material (CSAM). The technology is identical across the split; consent and disclosure are the entire dividing line.
Because the same diffusion and video models that draw a product shot can draw a nude, "AI porn" is less a distinct tool category than a use of general-purpose generators — sometimes on purpose-built adult platforms, sometimes by jailbreaking or misusing a mainstream one. That is exactly why it became a policy flashpoint: the capability is diffuse, cheap, and hard to contain, and the harm from the non-consensual end is severe and disproportionately aimed at women and girls.
The word "deepfake" itself was born in porn: in late 2017 a Reddit user of that name posted face-swapped explicit videos of celebrities, and the technique spread from there. The scale of the problem was quantified early — a widely cited 2019 report from Deeptrace (later Sensity) found 14,678 deepfake videos online, of which about 96% were pornographic and effectively all of those depicted women, most of them non-consensually. That figure is dated and predates modern diffusion models, but it set the framing that has held since: the overwhelming real-world use of face-swap AI was, and largely remains, non-consensual sexual imagery.
The 2022–2024 diffusion boom removed the last barriers to entry. Text-to-image models made a photorealistic nude generatable from a sentence, and "nudify" apps packaged the abuse into a one-tap consumer product. Enforcement and law then accelerated through 2025–2026. In the US, the TAKE IT DOWN Act — criminalizing knowing publication of non-consensual intimate images, including AI deepfakes, and requiring covered platforms to remove them within 48 hours of a report — was signed on May 19, 2025, with platforms given until May 19, 2026 to stand up their takedown systems. In August 2025, xAI's Grok Imagine launched a "spicy" mode that reporters showed would generate non-consensual nude video of Taylor Swift from a benign prompt, a vivid example of a mainstream tool shipping the capability with weak guardrails. In October 2025, OpenAI's Sam Altman said ChatGPT would allow erotica for age-verified adults as age-gating rolled out more fully. That plan slipped past its original December 2025 target, and in March 2026 OpenAI paused the adult-mode launch indefinitely, according to the Financial Times, as it refocused on coding and enterprise users; no erotica mode had shipped as of that pause. In the EU, the AI Act's Article 50 transparency rules — which require most deepfakes to be labeled — began applying on August 2, 2026, alongside an AI Act amendment, approved by the European Parliament in June 2026, that bans AI systems generating non-consensual intimate imagery or CSAM from December 2, 2026.
| Platform | Behavior |
|---|---|
| Nudify apps (Apple / Google stores) | The most-abused consumer form: upload a photo of a real person and the app generates a fake nude. Both Apple and Google prohibit these in policy, yet reporting through 2026 found such apps repeatedly slipping back onto both stores and advertising heavily on social platforms — the clearest example of enforcement lagging the capability. |
| xAI Grok Imagine | Launched a "spicy" NSFW mode in August 2025. Reporters demonstrated it would generate non-consensual nude video of a real named celebrity from a non-explicit prompt, making it the highest-profile case of a mainstream generator shipping explicit output with guardrails that failed on the exact category law now criminalizes. |
| OpenAI ChatGPT | Announced the opposite posture in October 2025 — Sam Altman said ChatGPT would permit erotica in text for age-verified adults, framed as a "treat adult users like adults" policy. The launch slipped past its original December 2025 target, and in March 2026 OpenAI paused the plan indefinitely, according to the Financial Times; it had not shipped as of that pause. |
| Meta (Instagram / Facebook) | Prohibits non-consensual intimate imagery and requires disclosure of realistic synthetic media. Enforcement leans on hash-matching and detection; accounts distributing deepfake porn or advertising nudify tools are removed, and the TAKE IT DOWN Act obliges rapid takedown on report. |
| Fanvue / OnlyFans-style platforms | The monetization endpoint for consensual, disclosed adult AI content — some now require AI personas to be labeled as such. This is the legitimate lane: a performer generating or authorizing their own synthetic content, as opposed to a bad actor fabricating someone else's. |
| Dedicated adult-AI generators | Purpose-built sites offer explicit text-to-image and video generation. Reputable ones enforce age verification and block real-person likenesses and CSAM; the gray-market end does neither, which is what regulators and payment processors are increasingly moving against. |
The honest way to think about AI porn is to stop arguing about the technology and start with consent, because that is the only axis the 2026 law actually cares about. A generator that can draw a nude is the same generator that draws a product mockup; the model has no opinion. What separates a compliant adult creator scaling their own authorized content from a criminal running a nudify app is not the tool — it is whether the person depicted said yes, and whether the audience is told what they are looking at. Every serious rule that landed this cycle, from the TAKE IT DOWN Act to the EU AI Act's labeling regime, encodes exactly that: consent and disclosure.
For mainstream creators and brands, the practical takeaway is a boundary, not a workflow. A governed content engine like Kompozy sits firmly on the safe-for-work, brand side of this line — it generates disclosed persona and avatar video, images, and posts for legitimate marketing, and explicit or non-consensual material is out of scope by design. The relevant lesson from the AI-porn story is defensive: the same face-lock and persona-video technology that powers a compliant AI thirst trap or a disclosed brand character can be abused to fabricate someone, which is why creators should understand likeness-detection tools and treat disclosure as non-negotiable. Know where the line is, stay on the right side of it, and label what you make.
A real AI avatar video from the Kompozy workspace. Kompozy makes disclosed, safe-for-work persona and avatar video for brand marketing.
Best for:
Not ideal for:
AI porn is sexually explicit media — images, video, voice, or text — generated or heavily altered by AI rather than filmed. It ranges from consensual, disclosed synthetic content (including fully fictional characters or media a performer authorized of themselves) to non-consensual material such as celebrity deepfakes and "nudify"-app images of real people made without their permission. The technology is the same across that split; consent and disclosure are the dividing line.
It depends on consent and subject. Consensual, disclosed adult content between the creator and the person depicted is generally lawful where regular adult content is. Non-consensual intimate images of a real person are increasingly criminalized — the US TAKE IT DOWN Act, signed May 19, 2025, criminalizes knowingly publishing them, including AI deepfakes. AI-generated child sexual abuse material is illegal everywhere with no lawful use.
Nudify apps take an uploaded photo of a clothed person and use AI to generate a fake nude of them. They are non-consensual by design, overwhelmingly used against women and girls, and banned in policy by both Apple and Google — though reporting through 2026 found them repeatedly returning to both app stores and advertising on social platforms. Using one to depict a real person is unlawful in many jurisdictions.
A US federal law signed on May 19, 2025 that criminalizes knowingly publishing non-consensual intimate images, including AI-generated deepfakes, and requires covered online platforms to remove such content within 48 hours of a valid report. Platforms were given until May 19, 2026 to establish the required notice-and-takedown process. It is the clearest US legal line drawn around non-consensual AI porn.
Policies diverge and shift. In October 2025 OpenAI said ChatGPT would permit erotica in text for age-verified adults — but the launch slipped past its original December 2025 target and in March 2026 OpenAI paused the plan indefinitely, and it had not shipped as of that pause. xAI's Grok Imagine shipped a "spicy" image and video mode in August 2025 that reporters showed would produce non-consensual nude deepfakes of a named celebrity, exposing weak guardrails. Most mainstream image and video tools prohibit explicit and non-consensual output, but enforcement is uneven.
That figure comes from a 2019 Deeptrace/Sensity report that found about 96% of the deepfake videos it catalogued were pornographic, nearly all depicting women non-consensually. It is a real, influential statistic but it predates modern diffusion models and today's volumes, so treat it as historical framing for how skewed the early use was, not a current measurement.
We checked every figure, date and platform rule on this page against the sources listed below on 2026-10-23, and rewrote or removed anything we could not confirm. Legal dates come from the FTC and the EU AI Act text. We updated the EU nudifier ban and OpenAI's paused adult-mode plan to match what has been reported.
Sources