deepfakeporn.wiki The independent reference on synthetic intimate imagery — technology, law, harm, recourse.
Reference → FAQ

Deepfake Porn: Frequently Asked Questions

DRBy Daniel Reyes · Last updated: November 2026

The ten questions readers ask most, answered at reference depth. Each answer links onward into the sections of the site where the topic gets full treatment.

Q1.Is deepfake porn illegal?
In most developed jurisdictions, yes — creating or sharing sexualized deepfakes of a real person without consent is now unlawful. The UK criminalizes creation itself; the US federal Take It Down Act criminalizes non-consensual publication and forces platforms to remove reported material within 48 hours; roughly 45 US states and most EU members have their own statutes. Where the subject is a minor, it's prosecuted as child sexual abuse material. Full map on the law & safety page.
Q2.How is deepfake porn made?
Four pipelines dominate: inpainting (a clothing mask on a real photo is regenerated by a diffusion model), identity-conditioned generation (LoRA fine-tunes or face-embedding adapters), face-swap video, and text-to-video generation. The first needs only one clothed photo. Details in the technology guide.
Q3.Can deepfake porn be detected?
At platform scale, usually; at frontier quality for a single image, no. Forensic classifiers find artifacts — inpainting boundaries, generator fingerprints, swap-edge discontinuity — but accuracy falls with each model generation and under re-compression. The durable countermeasures are provenance signing of real media (C2PA) and hash-matching of known material (StopNCII), not classifier verdicts.
Q4.What should I do if someone made a deepfake of me?
Preserve evidence first (URLs, timestamps, screenshots). Report via the platform's NCII channel — covered platforms must act within 48 hours under the Take It Down Act. Fingerprint the image with StopNCII (adults) or Take It Down (minors) so re-uploads get blocked. File search-engine removal requests. Then police report and counsel. The full sequence is in section 8 of the reference.
Q5.Why do these sites keep disappearing and reappearing?
Domain churn. Registrars, hosts, CDNs, and payment processors all eject AI-NCII services, so operators treat domains as disposable — rotating TLDs and respelled brands on a weeks-to-months cycle while the Telegram channel remains the stable identity. See the ecosystem section.
Q6.Is private creation — without sharing — illegal?
Jurisdiction-dependent and tightening. The UK criminalized private creation in 2024; several US states criminalize it with intent elements; elsewhere it may be civil rather than criminal. It also exposes the creator: uploaded photos live on the operator's servers, and tool databases have already leaked. Covered in the safety guide.
Q7.Do deepfake porn tools actually work?
The technology is real — inpainting output can be convincing because most pixels are the victim's real photo. What routinely fails is the market: many sites are outright scams, most retain uploads, and the rest churn through domains before delivering consistent service. See the comparison page.
Q8.Who gets targeted?
Overwhelmingly women. Early waves hit celebrities because public photo libraries feed identity conditioning well; since 2023 the majority of victims are private individuals — classmates, partners, streamers, colleagues — processed via ordinary social-media photos.
Q9.Can I tell if a nude image is fake by looking?
Sometimes, unreliably. Diffusion anatomy tends toward airbrushed symmetry; inpainted regions are too smooth; face swaps show jawline seams and drift under occlusion. Missing provenance metadata is itself a signal. Frontier output defeats eyeballing — forensic tooling is needed for anything consequential.
Q10.What is being done about it globally?
A three-layer response is converging: criminal statutes (US federal + ~45 states, UK OSA, EU directive, Korea, Australia), infrastructure enforcement (payment-processor bans that shrank the commercial tier), and prevention technology (C2PA signing, StopNCII/Take It Down hash-blocking). Open-weight models remain the unsolved layer.