The Undress AI Index A data desk on the synthetic-intimate-imagery market

Data briefing · November 2026

Undress AI, measured

Forty domains, three technologies, one legal trendline — the "undress ai" market as a dataset.

By Henrik Dahl, data editor · Last updated: November 2026

"Undress AI" is simultaneously a brand family, a keyword and an industry. Measured as a dataset — domain registrations, enforcement records, pricing samples and takedown counts — it resolves into a clear structure: a high-churn market where brand names rotate faster than the underlying models change, where the legal environment hardened dramatically between 2023 and 2026, and where the only reliable numbers victims can act on are removal deadlines. This briefing presents the current figures. Everything here is directional, sourced, and footnoted — methodology is on the methodology page.

40+
distinct undress-ai-branded domains observed in WHOIS history since 2023¹
~4 mo
median brand lifespan before domain rotation or suspension²
~90%
of services paywall usable output behind credits or crypto
48 h
statutory removal deadline under the US TAKE IT DOWN Act (2025)
48+
US states & DC criminalizing nonconsensual intimate-image sharing
~96%
of deepfake video online is nonconsensual intimate imagery (2019–24 studies)³

Fig. 1Domain churn: the defining metric

The single most diagnostic statistic in this market is not output quality or pricing — it is registration turnover. Tallying the undress-ai-branded domains appearing in WHOIS records per year shows a market that accelerates even as its brand names vaporize:

0102030 82023 172024 222025 17*2026 YTD
Fig. 1 — Distinct undress-ai-branded domains first observed per year. *2026 through November. Note the 2026 dip coincides with TAKE IT DOWN enforcement, not reduced demand — operators shifted to unbranded churn domains. Source: WHOIS timeline sampling.¹

Two readings from that chart matter. First, the market grew through every enforcement wave — demand is inelastic to domain loss because operators treat registrations as disposable. Second, the apparent 2026 plateau is an artifact: after federal removal duties took effect, operators abandoned keyword-branded names for random-string domains, which our sampling method undercounts.

Fig. 2Market structure, quantified

Table 1 — The undress ai market as measured (November 2026 sample)
MetricValueBasis
Price per image (paid tiers)$0.10 – $5.00Advertised credit pricing across 30+ observed services
Operator GPU cost per image~$0.02 – $0.10Spot-instance inference estimates, SDXL-class inpainting
Free-tier prevalence~70% of servicesOffer at least one free/watermarked render — uploads are the monetization
Payment methods~60% crypto-inclusive; ~40% cardCard acceptance declining as processors enforce policy
Claimed retention ("auto-delete")~85% of servicesAuditable in 0% — no observed service offered proof
Median time-to-paywall1 imageFree tier = one watermarked sample, then credits

Fig. 3The demand side, estimated

What powers that supply is measurable too — approximately. Triangulating keyword-tool volume estimates, autocomplete suggestion counts and referral observations gives a demand picture that is large, stable and mostly informational-criminal in intent:

Table 2 — Demand-side indicators for "undress ai" and adjacent keywords
IndicatorEstimateTrend
Monthly global search volume (keyword family)~1–2MFlat-to-up since 2023; resilient to delisting
Share of searches that are brand-navigational~35–45%Falling as brands churn faster than memory
Share arriving via Telegram/shortlink referrals~25–35%Rising — the churn adaptation
Peak demand periodsNews-cycle spikes (±200–400%)Each enforcement story recruits new demand

The last row deserves emphasis: enforcement publicity functions as acquisition marketing. Every takedown story produces a measurable search spike — a Streisand curve that the market's churn model absorbs by design. It is the structural reason suppression strategies focus on the payment and hosting rails rather than on names.

Fig. 4The enforcement trendline

Against that market data sits a legal dataset moving in the opposite direction. The count of operative legal instruments reached by November 2026:

The full legal dataset, with penalty ranges, is tabulated on the legal risk page.

Fig. 5What victims can act on — the useful numbers

For the dataset's most important users, three figures outrank all others: 48 hours (federal platform-removal deadline), 24–72 hours (typical hash-match propagation through StopNCII's partner network), and 1 photo (minimum input modern identity pipelines need to produce a passable composite). The removal-route table on the legal risk page scores each option by measured speed.

One more number frames all the rest. Of the forty-plus branded domains we have tracked since 2023, the overwhelming majority are dead or rotated as of this writing — while the statutes enacted in the same period remain in force everywhere they passed. Infrastructure churns; law compounds. That asymmetry, more than any single figure, is what the 2026 dataset is ultimately about.

Method note — Domain counts derive from WHOIS timeline sampling of undress-ai-branded registrations; market structure from advertised pricing sampled November 2026; enforcement counts from statute texts and legislative trackers. Figures are directional estimates for a deliberately opaque market — confidence intervals and limitations in the methodology.
Next in the series: the methods page numbers the technical pipeline; legal risk tabulates the statutes; measured quantifies each service archetype; the FAQ answers the most-queried numbers.

¹ Domain counts: WHOIS/RDAP registration-history sampling, deduplicated by operator fingerprint where linkable; unbranded churn domains undercounted by design.

² Median lifespan: first-observed to last-resolved-or-content-abandoned, branded domains only.

³ Share-of-content estimate: synthesis of Deeptrace (2019) and successor studies through 2024; the ratio has remained stable even as absolute volume grew.