ClothOff.brief
Home › Guide

How ClothOff Worked — The Quick Guide

By Danny Mercer · Last updated: November 2026 5 min read

TL;DR

ClothOff ran a four-step pipeline: upload → clothing mask → diffusion inpaint → store. Simple on the surface. The dangerous step was the invisible one: every uploaded photo of a real person went into operator storage with no enforceable deletion promise.

Part 1The pipeline, step by step

Step 1 · Upload. User sends in a photo — usually of a real, clothed person. The server now holds that photo. This is the consent failure point, and it happens before any AI runs.
Step 2 · Segmentation. A pose- and garment-detection model marks which pixels the generator may change — the clothing region. Mask quality is what separates convincing tools from sloppy ones.
Step 3 · Inpainting. A diffusion model regenerates the masked area, guided by body-prior training. Face, hair, and background are deliberately preserved — recognizability is the feature.
Step 4 · Deliver & retain. Output goes back to the user. Input and output stay on operator infrastructure — under privacy policies that typically claim a broad license and no retention limit.

Part 2Why it went viral

Three ingredients. Free previews lowered the barrier to zero. Output quality crossed the "believable at a glance" threshold around 2023. And Telegram bots made distribution frictionless — no app store to get kicked out of, shareable in a forward.

The virality wasn't accidental. The product was built for the single most shareable premise imaginable, and the referral loop — see a result, send it on — did the marketing.

Part 3The whack-a-mole era

Then enforcement started landing. App stores pulled the mobile apps. Registrars acted on abuse complaints. Card processors, whose acceptable-use rules bar NCII services, cut payment access. The operators' answer was churn: new domain, same backend, repeat. When the web kept burning, the brand shifted weight to Telegram bots, which replicate faster than registrars can move.

Net effect: today "ClothOff" is a name dozens of clones wear. Some run the original stack; many are copycats; some are pure phishing wearing the brand for traffic. There is no official destination to check, which is itself the tell — a legitimate service doesn't need to exist in twelve places at once.

Quick read — what this means

Part 4The five myths, sixty seconds each

Part 5The part nobody saw: storage

Security researchers looking at this niche keep finding the same thing: upload buckets left open, user photo libraries browsable, datasets resold. ClothOff-era services were no exception to the pattern. A photo uploaded "just to try it" enters storage the uploader can't audit and can't recall — and the person in the photo never consented to any of it.

That's the real mechanism to understand. The AI gets the headlines; the retention does the lasting damage.

And one practical consequence follows: if a photo of you was ever processed by any service in this family, the damage control isn't deletion requests to an anonymous operator — it's the blocking tools. Registering real photos with StopNCII or NCMEC's Take It Down hashes them on-device and stops platforms accepting re-uploads, whether the original service still exists or not. That route is detailed in the safety brief.

Part 6Why nobody could just "shut it down"

Fair question, and the answer teaches the niche. There was never a single switch: the code was portable, the brand was a name anyone could wear, and every enforcement win — a suspended domain, a pulled app, a dropped processor — targeted one copy at a time. Regulators and researchers could document and disrupt; they couldn't delete a replicable pipeline. That asymmetry is the reason this brief reads like a field guide instead of an obituary.

Part 7The vocabulary, in 30 seconds

Three terms cover everything you'll read about this niche. Inpainting is the technique — mask a region, let the diffusion model fill it. Undress / nudify is the product label — inpainting applied to the clothing region of a real photo. Domain churn is the market behaviour — rebranding across disposable domains under enforcement pressure. If a new term appears tomorrow, it's a new label on one of those three.