Pet-portrait apps hiding rape-scene generators: a 400-app deepfake network scaled through Meta ads

More than 400 connected apps and websites posing as pet-portrait generators and other everyday tools served explicit deepfake interfaces, including templates simulating rape, to users who arrived the right way, and the network behind them scaled for roughly a year on Meta's ad engine before Apple and Google pulled the listings 1. WIRED's investigation, with the nonprofit AI Forensics and security firm Infoblox, traces the operation to Hong Kong-registered KeLing Technology Limited: apps paid for through Apple Pay and Google Pay, more than 350,000 downloads counted on Google's Play Store alone, and distribution bought with tens of thousands of ads on Facebook and Instagram 1. Apple removed 19 apps after WIRED shared the associated domains; Google had already pulled the majority before being contacted 1.

The takedown count is the wrong ledger. The more-than-350,000 figure covers only the 27 network apps still live on the Play Store in September 1; DiffAI, created December 2025, had itself been downloaded more than 100,000 times from the Play Store, more than a quarter of the entire counted figure across the 27 September-listed apps, on the printed figures 1. Concentration like that is what a funnel looks like: many disposable doors, a few that actually convert, and a factory behind them minting roughly 30 new app-and-website pairs a week at the peak 1. The apps are the disposable layer. The asset is the pipeline above them.

Storefront review approves a shell the server rewrites

The pet-portrait app Viefur sells about $4 for 130 coins to dress pets in human clothing; under conditions its developers control, the same binary loads a web interface of pornographic deepfake templates, scenes named "bedroom rape" and "threatening knife" among them, where users insert photos of real people 1. When downloaded, each app reports where the user is coming from, the language and a unique identifier to a server that replies with the benign or the explicit version; one apparent trigger is arriving from a Meta ad click, found AI Forensics lead researcher Paul Bouchaud, who reverse engineered the code 1. The binary the store reviewed is a costume; the product lives on a server that flips after approval.

Apple's own accounting says the costume trick is an industrial pattern, not one vendor. "We review every app before it's published, and these apps complied with our rules when we reviewed them," Apple spokesperson Adam Dema told WIRED, explaining that developers later swapped the website the app pulls content from 1. Apple says it has suspended tens of thousands of apps for these "bait-and-switch" tactics over the past year and recently tightened store rules on "nudify" tools 1. And the operator kept shipping through the pressure: in late September, users of the explicit interface received a notification announcing a new AI model, Veyra, taglined "More real. More natural. More wild," with a homepage contact address belonging to KeLing 1. About a month before the mass removals, the network was still doing product launches.

The whole fleet documents itself. The network's websites carry a JavaScript bundle listing the domains, app package details, encryption keys and Facebook ad trackers of the operation, the configuration trail Infoblox staff threat researcher Maël Le Touz used to identify almost 500 linked domains running the same code 1. Delete a listing and the configuration survives to seed the next clone.

The ad engine was the real storefront, and it stayed open

Distribution ran through Meta under apparently fabricated payer names such as Draftmondcreldbro Duhustepre, published in Meta's own ad disclosures alongside pages with no followers 1. Meta says it banned KeLing Technology Limited from advertising in 2025 after the company failed to comply with a cease-and-desist letter, and that it has kept trying to advertise "through other ways" 1. The network's registrations nonetheless spiked the following spring 1. After WIRED handed over the 400-plus domains, Meta reported more than 3,000 associated ads, two-thirds already removed, and enforcement against 11,600 ad accounts, more than 4,700 business pages, 5,200 users, 15,000 pages and 209 apps blocked from advertising 1. Every line item counts artifacts, and manufacturing artifacts is this operation's core competence.

The published reporting does not price KeLing's own ad spend; the visible economics are Meta's. Using payer names that the EU's Digital Services Act forces Meta to disclose, the Tech Transparency Project tied 30,800 face-swap and AI-app ads across 210 Facebook pages to GatherOne, one of the authorized Chinese reseller agencies that sell Meta advertising, with more than 7,600 of them promoting apps TTP verified could produce nonconsensual nudes 2. Reuters, as cited by TTP, reported Meta's internal estimate that about 19 percent of its China ad revenue, more than $3 billion, came from banned-content ads 2. In January 2026, GatherOne co-founder and CEO Wang Hao posed with Meta's top China executives holding one of four industry performance awards the company collected at Meta's Agent Summit in Hong Kong 2. Those are not KeLing's ads; they are the distribution channel this network rode, with receipts.

A year of enforcement, read in order

No single source prints this sequence. Assembled across five outlets and research organizations, it is the story:

• January 2025: 404 Media reports that, per Similarweb data, the nudify service Crushmate drew 90 percent of its roughly 240,000 visitors from Facebook and Instagram ads 3.
• 2025, month unstated: Meta bans KeLing Technology Limited after a failed cease-and-desist; the network's first apps appear toward the end of the year, DiffAI among them in December 1.
• June 2026: after a TTP investigation, Apple and Google block storefront searches for "nudify," "undress," "deepnude" and "deepfake"; app names like Viefur and DiffAI never carried those terms 4.
• July 2026: TTP identifies GatherOne behind 30,800 AI-app ads; GatherOne tells TTP it has suspended new nudify-related ad account applications 2.
• Early September 2026: AI Forensics says it discovered the operation behind more than 400 nudifier apps on Google Play and the Apple App Store; Danish journalists report portions of the network; 27 apps remain live on the Play Store 51.
• Late September 2026: the network announces its Veyra model to users of the explicit interface 1.
• November 2, 2026: WIRED publishes; Apple removes 19 apps on contact, Google already has the majority down; after WIRED's outreach, connections to KeLing were scrubbed from associated channels and sites, and similar ads and apps are still being published as the story closes 1.

Meta identified the payer in 2025 and banned it; the apps kept shipping, peaked in the spring, and were still launching models in September 1. "They don't expect enforcement to catch up, and so far it hasn't," Bouchaud told WIRED 1. Enforcement keyed to artifacts, one app, one ad, one search term at a time, loses weekly against infrastructure that re-clones in days. The surfaces this network cannot cheaply regenerate are the ones it never owned: payer identity, payment rails, registrar records, and those sit with the platforms. AI Forensics published its findings inside what it calls a one-month countdown to the EU banning "nudifier" apps 5. A ban on the apps is enforcement against the inventory; the distribution and the till stay open for the next operator.