TMCompiled by Theo Marsh · Last updated: November 2026
The ten questions that arrive at the desk most often, answered from the aggregated consensus rather than one writer's opinion. Deeper treatment lives on the pages each answer links to.
R.1What is an AI undress app?
A service that converts a clothed photo of a real person into a synthetic nude — a specialized inpainting pipeline that masks clothing and regenerates that region while the victim's real face, pose, and background pass through untouched. The main brief defines the category.
R.2Is using one legal?
On a non-consenting person: criminal at creation in the UK and Korea, criminal at sharing nearly everywhere, civilly actionable in most US states, GDPR-violating in the EU. On a minor: CSAM prosecution everywhere. Consent-verified self-use is the narrow lawful lane — see the safety consensus.
R.3Do they actually work?
Yes — the consensus bar sits at 96% because most output pixels are the real photograph. What fails is the service layer: scams, dying domains, retained uploads.
R.4Where does my uploaded photo go?
Onto the operator's retention store under terms granting a perpetual license — into retraining corpora and, in documented cases, into leaked databases alongside users' payment details. Zero community disagreement on this point.
R.5Illegal if I never share it?
Jurisdiction-dependent and tightening: creation is criminal in the UK and Korea, criminal-to-civil-to-unaddressed across US states. The upload itself — someone's photo shipped to an offshore server — carries exposure regardless.
R.6Why do they keep changing domains?
Infrastructure ejection: registrars, hosts, CDNs, and payment processors all ban the category, so domains rotate on a weeks-long cycle while the Telegram channel is the durable identity. Community rule: any review older than a month describes a dead service.
R.7What do the tools' own users warn about?
Their own advice, verbatim themes: assume your upload will leak eventually; free tiers exist to harvest photos; the domain you paid last month is probably gone; the legal risk is personal; nobody has ever gotten a refund. The position map locates the user bloc in full.
R.8Can the results be detected?
At platform scale yes — regenerated regions carry statistical artifacts and lack provenance credentials. For a single frontier-quality image, detection is unreliable; provenance signing of real media is the durable countermeasure.
R.9Someone undressed my photo — options?
The validated sequence: preserve evidence → platform NCII report (48-hour US duty) → StopNCII/Take It Down fingerprinting → search removals → police report + specialist support. Ordered detail on the safety page.
R.10What's still actually disputed?
Two things: whether private creation without sharing should be criminal (~58% bar — the genuinely contested question), and whether open-weight models can be governed at the model layer (~41% — the researchers' open problem). Everything else in this brief is settled.