Undress app
An undress app is a software application — typically delivered as a website, Telegram bot, or mobile-adjacent service — that accepts a clothed photograph of a real person and returns a machine-generated image of that person depicted nude or partially nude. The resulting image is synthetic: the software generates plausible anatomy rather than recovering anything hidden in the photograph. The term "undress app" is the most common consumer phrasing for the category in English-language search, ahead of "nudify" and "undress AI".
undress app (noun) — a class of image-manipulation services that fabricate intimate imagery of identifiable persons from clothed photographs, using generative models (predominantly diffusion inpainting since 2022). Distinguished from general photo editing by intent and output, not by underlying technique.
1. Definition and scope
The label covers any service whose marketed function is the fabrication of intimate imagery of real people from ordinary photographs. Three properties delimit the category: (a) the input is a real, identifiable person's photograph; (b) the depicted act — undressing — is applied without that person's involvement; (c) the output is presented as an image of the subject rather than as an illustration. The category is a specialized branch of what the wider literature calls deepfake media, and its legal treatment tracks the broader concept of non-consensual intimate imagery.
1.1 Boundary cases
Tools that generate fully synthetic persons, stylized avatars, or licensed model imagery fall outside the category as defined here — the decisive variable is whether a real, identifiable person is depicted. Edge cases include face-swap pornography (a related but distinct technique that grafts a face onto an existing explicit image rather than synthesizing a body) and "age progression" or "body editing" tools that share the same pipeline but stop short of nudity; enforcement practice tends to treat them under the same statutes where the output is sexualized.
1.2 Why the definition matters
Statutes and platform policies regulate the depiction, not the tool. A mainstream editor capable of the same inpainting is not an "undress app" in enforcement practice; a one-button service that does nothing else is. This is why the category can be documented — and prosecuted — as a category: intent is legible in the marketing, the single-purpose interface, and the checkout flow.
1.3 Naming across search and press
The same services surface under "undress app", "undress AI", "nudify", "clothes remover", and the legacy "deepnude" label. Operators exploit the ambiguity deliberately: a burned keyword is abandoned while an adjacent spelling is registered, so search results for the category churn continuously. This wiki uses "undress app" as the headword because it remains the highest-volume phrasing.
- "Undress app" / "undress AI" — the consumer-facing phrasing; dominates English search volume.
- "Nudify" — the press and research literature's preferred label (Graphika, Sensity).
- "Deepnude" — the legacy headword from the 2019 app; now mostly a search artifact and a brand signal for clones.
- "Clothes remover" and adjacent spellings — keyword-shape variants that exist mainly because the primary terms are penalized.
The naming layer matters practically, not just lexically: regulators, journalists, and victims searching under different words find different slices of the same supply, which is one reason the category looked smaller than it was for as long as it did.
2. History
2.1 The 2019 proof of concept
The desktop application DeepNude demonstrated the pipeline on a pix2pix-class generative adversarial network. Press coverage in June 2019 led the author to withdraw it within days; leaked copies circulated regardless, and the episode fixed the category's business pattern: a single-purpose tool, priced access, instant notoriety, immediate suppression — followed by clones.
2.2 The 2020–2021 bot migration
Researchers documented Telegram bots automating the workflow at scale; one Sensity investigation measured six-figure volumes of generated images, the large majority synthesized from ordinary social-media photographs of private individuals. The bot era proved two things that shaped everything after: demand was mass-market rather than fringe, and the victims were overwhelmingly women who had never interacted with the service. Payment during this period migrated to crypto micro-transactions, establishing the pattern — card networks terminate, operators switch rails — that still governs the category's checkout pages.
2.3 The diffusion era, 2022 onward
Open diffusion models plus fine-tuning adapters raised output quality sharply and cut per-image cost to cents on rented GPUs. The commercial form stabilized into credit-based web fronts deployed across rotating keyword domains, with Telegram distribution as the low-friction parallel channel. Output quality reached the point where results pass a casual glance — which shifted public attention from the technology to the harm.
2.4 The legislative response, 2023–2026
Law enforcement and legislatures responded between 2023 and 2025: the United States passed the TAKE IT DOWN Act; the United Kingdom extended the Online Safety Act regime to cover creation as well as sharing; most US states amended NCII statutes to name synthetic imagery; the EU's Violence-Against-Women directive began transposition across member states. Platform enforcement ran in parallel — app-store purges, ad-network severances, and the 2024 San Francisco suit against sixteen operators' services. See the safety and law entry for the jurisdiction table and the removal pathway.
3. Technology
3.1 The four stages
The canonical pipeline has four stages: photo intake, person/clothing segmentation (masking), diffusion-based inpainting of the masked region, and post-processing (tone matching, re-graining, upscaling). The first and last stages are conventional computer vision; the middle two are where the fabrication happens.
- Intake — the upload enters operator infrastructure; the riskiest hop for the subject of the photo.
- Masking — segmentation and pose estimation mark the clothing region.
- Generation — the diffusion model fills the mask with invented anatomy.
- Composite — tone matching, re-graining, and upscaling blend the region back into the photograph.
3.2 Why the output is fabrication, not recovery
The model synthesizes anatomy statistically — it is not recovering hidden data, because no such data exists in the pixels of a clothed photograph. What the model returns is a prior-based guess conditioned on visible silhouette and skin tone. Consequently two runs on the same input produce different bodies, and identifying details beneath the clothing — scars, tattoos, exact proportions — are always invented. This property matters legally (the image is a depiction, not a disclosure) and forensically (erased identifying details are the strongest authenticity signal).
3.3 Characteristic failure modes
Output quality depends on pose simplicity, occlusion, and input resolution, and fails in characteristic ways: erased skin details, warped anatomy at occlusion boundaries, pattern bleed from busy clothing, and a detectable synthetic texture ("sheen") inconsistent with the photograph's sensor noise. The detection literature — the FaceForensics++ line of work and its successors — classifies on exactly these artifacts, which is why fabricated imagery remains reliably identifiable even as consumer-facing quality improves. A fuller treatment is in the guide entry.
4. Distribution channels
- Credit-based web fronts
- The dominant form: a landing page, an upload box, a credit paywall. Frequently deployed in clusters — one operator, many domains — which produces the category's signature domain-churn pattern.
- Messaging-platform bots
- Telegram hosts the largest volume. Zero-friction upload, crypto micro-payments, and near-zero moderation; uploads commonly recirculate within bot channels.
- Local/open-source builds
- Public model weights and inpainting code allow self-hosted generation. Removes the operator from the data path but not from the legal analysis.
- Reseller APIs
- Wholesale inference backends let affiliates launch branded fronts quickly — the structural reason clone sites share behavior and appearance.
4.1 Why the churn is structural
Every channel except local builds depends on infrastructure that can be revoked: registrars suspend domains, hosts terminate accounts, card networks cut processing. Operators respond by treating each domain as disposable — a fresh name resets negative search results, blacklist entries, and accumulated scam reports. The comparison entry maps the storefront shapes this produces.
- Spelling pivots — dropped vowels and appended letters; domains that exist only because the dictionary word is burned.
- TLD migration — wholesale moves from .com/.net reputation systems into cheap new gTLDs with thinner abuse desks.
- Template reuse — identical page structure, copy, and policy text redeployed under the new name, which is how churn is measured.
5. Legal status
By late 2026 the category sits inside overlapping prohibitions rather than a grey area — the question a reader should bring to any specific service is no longer "is this legal" but "which instrument applies first."
5.1 United States
The TAKE IT DOWN Act (2025) criminalizes nonconsensual intimate imagery including digital forgeries and obliges covered platforms to remove within 48 hours of a valid report; 15 U.S.C. § 6851 provides a civil action; and the large majority of state statutes — Virginia § 18.2-386.2, California's SB 926 and AB 602, Texas SB 1361 among them — expressly cover fabricated depictions.
5.2 United Kingdom and European Union
In the UK, sharing is criminal under the Online Safety Act 2023 and a 2025 provision added a creation offence; the UK Revenge Porn Helpline handles synthetic-image cases routinely. In the EU, victims hold GDPR erasure rights and DSA notice-and-action against hosting platforms, while member-state criminal codes supply the offence provisions. Where the depicted person is a minor, child-protection statutes apply everywhere with far heavier penalties — the National Center for Missing & Exploited Children is the US clearinghouse.
5.3 The consent principle underneath
Every governing instrument reduces to one test: a real person, an intimate depiction, an absence of consent. The falsity of the image is not a defense — modern statutes regulate the depiction itself. The Electronic Frontier Foundation maintains running analysis of where these statutes overreach and where they bind; the Cyber Civil Rights Initiative tracks the victim-side experience of the same laws. The practical consequence for the would-be user is blunt: "it was just a test" and "I never shared it" are the two fact patterns the newer creation-side statutes were written to reach.
6. Documented incidents
6.1 School and community cases
- Westfield, New Jersey (2023). High-school students fabricated images of classmates; the case accelerated the state's legislative response and national school-policy debates.
- Almendralejo, Spain (2023). Undress-app images of schoolgirls circulated locally; prosecutions followed under child-protection law.
- Beverly Hills (2024). Middle-school students expelled after fabricated images of classmates spread — one of several US school incidents that year.
6.2 Enforcement waves
- San Francisco suit (2024). The city attorney sued operators of sixteen of the most-visited undressing sites — the first US municipal action aimed at the services rather than individual users.
- Platform enforcement (2024–2025). App stores, ad networks, and payment processors removed or severed clusters of undress-app services; Graphika's 2024 research measured tens of millions of monthly visits to nudify services, motivating the crackdown.
- Statute era (2025–2026). Federal and creation-side offences produced the first charging activity against makers and uploaders, not only distributors.
Read together, the incident list carries the entry's recurring lesson: every documented case involved ordinary photographs of private individuals pulled from social media, and none of the subjects had any relationship with the service that depicted them. The harm was never a fringe edge case; it was the product's primary use.
7. Category economics
7.1 Why services persist
A replacement storefront costs a domain fee, a copied template, and a day of setup; ninety days of credit sales covers that several times over. Enforcement, by contrast, is measured in weeks per domain and split across registrar, host, and payment rails that do not coordinate. The asymmetry is the category's central economic fact: nothing in the last seven years has raised the marginal cost of launching a new domain above the revenue of running it briefly.
7.2 The evidence trail an upload leaves
Whatever the storefront promises, a single use produces records at several independent hands:
- The operator's logs — upload, account, and IP records on anonymous infrastructure.
- The payment rail — card or blockchain records that survive the merchant's shutdown and surface in discovery or breach dumps.
- The platform layer — CDN, registrar, and analytics identifiers that link the burned domain to its successors.
- The output itself — a fabricated image that, once circulated, carries the artifact signature described in the guide entry.
7.3 The scam layer
The same churn produces clones whose actual function is fraud: card skimming at checkout, credential harvesting, malware packaged as "cracked" versions, and extortion schemes that reuse uploads against the uploader. Scam reports of this shape route through the FTC's fraud portal and the consumer guidance at consumer.ftc.gov; platform-level image removal runs separately through Google's removal request flow.
8. Short answers
Is an undress app's output a real photo of the person?
No. The model synthesizes anatomy statistically; identifying details beneath the clothing are invented, not recovered. Two runs on the same input produce different bodies. The guide entry covers the mechanism.
Is using one on someone's photo illegal?
In most jurisdictions, yes where consent is absent — the US TAKE IT DOWN Act and most state NCII statutes, the UK OSA regime including the 2025 creation offence, and member-state law across the EU all reach fabricated intimate depictions. The safety entry tabulates it.
What should a victim do first?
Document URLs and screenshots, then hash-register the image at StopNCII (adults) or NCMEC Take It Down (under-18s), then file platform removal requests citing the 48-hour duty where applicable.
Why do the domains keep changing?
Registrar suspensions, hosting cutoffs, and payment-processor terminations force redeployment on fresh keyword domains — the structural churn described in section 7.
Does this wiki operate or link to these tools?
No. undressapp.wiki is a documentation project with a strict no-linking, no-endorsement policy; entries exist for reference, safety, and research — see the editorial policy.