Inside our similar-performer data there are 7,004 references that point to identities with no name attached. That is 52% of all 13,420 references in those lists — more than half of every "who looks like this person" pointer we hold leads to someone the published record never named (our index, 2026-08 snapshot).
This is the most honest number we can publish about our own data, and it is also the most useful one, because it quantifies something every face search tool has and none of them show you: the gap between faces that exist and faces anyone has bothered to identify.
What exactly does the number 7,004 count?
7,004 counts references, not people. Being precise about that is the point of this article.
Each identity record in our index carries a short list of its nearest neighbours in vector space — the faces the model considers most similar. Some entries in those lists resolve to a named performer profile. The rest resolve to an identity that exists only as an internal key, a representative image, and a vector.
| Quantity | Value | What it is |
|---|---|---|
| Total similar-performer references | 13,420 | Every entry across all neighbour lists |
| References to a named profile | 6,416 | 48% of the total |
| References to an unnamed identity | 7,004 | 52% of the total |
| Distinct unnamed identities referenced | 589 | The people behind those 7,004 pointers |
| Performers with at least one unnamed neighbour | 1,899 | Of 2,070 records that have a neighbour list |
| Identity records in the index | 2,333 | Of which 1,707 carry a resolvable named profile |
Source: our index, 2026-08 snapshot, n=2,333 identity records.
So: 7,004 pointers, 589 people. The average unnamed identity is somebody's near-neighbour about twelve times over. If you only read the headline number you would overstate the problem by roughly a factor of twelve — which is why we publish the breakdown next to it.
Why do faces end up in an index without a name?
Because the face and the name arrive through different pipelines, and only one of them is reliable.
A face enters the index whenever a detector finds one in an indexed image. That is mechanical and near-complete. A name enters only when some published page states it in a form we can parse and link. That is editorial, inconsistent across sites, and frequently absent entirely.
The common causes, in rough order of how often we see them:
- The source page credits no performer at all — very common on aggregator sites, which republish material with the metadata stripped.
- The page credits a name only in Japanese, in a form that never got mapped to a canonical profile. Of our 2,333 identity records, 626 exist as a raw name string with no linked profile and no biographical fields at all.
- The performer worked under multiple names and no source connects them.
- The face belongs to someone incidental to the production and was never a credited performer in the first place.
The last one is worth stating plainly: not every face in an adult video index is a performer, and an index that assumed otherwise would be wrong in a way that matters.
What does an unnamed neighbour tell you in a search result?
It tells you the resemblance is real and the identification is missing — which is different from a failed search, and much more common than users assume.
1,899 of the 2,070 records with a neighbour list (92%) have at least one unnamed identity among their nearest neighbours. If you run a face search and get back a result with a score above threshold but no name, you have not hit a bug. You have hit the actual state of the public record.
This is also why we do not filter unnamed identities out of results. Suppressing them would produce a cleaner-looking list that quietly misrepresents the data: it would turn "this person exists but was never credited" into "there is nobody here". The first statement is true and the second is not.
How does this shape the coverage numbers?
It drags them down, and we report the lower figures rather than the flattering ones.
Every coverage percentage we publish uses the full population of 2,333 identity records as the denominator — including the 626 that carry a name string and nothing else.
| Field | Records with a value | Coverage of n=2,333 |
|---|---|---|
| Similar-performer list | 2,070 | 89% |
| Cup size | 1,375 | 59% |
| Debut date | 1,371 | 59% |
| Height | 1,340 | 57% |
| Measurements | 1,316 | 56% |
| Birth date | 1,301 | 56% |
| Birthplace | 716 | 31% |
| Blood type | 547 | 23% |
Source: our index, 2026-08 snapshot, n=2,333.
We could raise every one of these numbers by quietly changing the denominator to the 1,707 records that have a real profile. Blood type would go from 23% to 32%, birthplace from 31% to 42%. We do not, because the 626 are in the index and they are returned in searches, so excluding them from the accounting would describe a database we do not actually run.
Data and method
Source. The starlikeness face index: face vectors and identity records derived from cover images and stills across 106 sites.
Sample. 241,792 face vectors; 2,333 identity records; 13,420 similar-performer references across 2,070 neighbour lists.
Method. Faces detected with YuNet, aligned from five landmarks, encoded with ArcFace into 512-dimension vectors. Neighbour lists are the nearest vectors by cosine similarity, subject to a 0.40 same-person threshold. A reference is classified as "unnamed" when it resolves to an identity keyed only by an internal hash, with no linked performer profile.
Snapshot date. 2026-08. All counts recomputed from the live data file on that date.
Known bias. Our sources are highly concentrated. 2,246 of 2,333 representative images come from a single site, and of the 106 sites we index, only around 13 carry Japanese content IDs at all — the rest are largely international tube and aggregator sites. The naming gap we measure is therefore partly a property of those sources' metadata habits, not of the industry as a whole. What this index measures is exposure on aggregators, not release volume. A well-documented performer whose work is absent from our sources would appear here as absent, and a poorly credited aggregator upload inflates the unnamed count.
What we cannot compute. The index stores faces, not titles. There are no release-date, studio or title-count fields, so no statement about production volume, studio market share, or year-on-year output can be derived from it.
Citation.
starlikeness (2026). "Our Index Contains 7,004 Faces With No Name."
Face index, 2026-08 snapshot (n=2,333 identity records; 13,420 neighbour
references). https://starlikeness.com/en/posts/faces-without-names
Related questions
- How much of the industry can we actually see?
- Where our data comes from, and what's missing
- What does a similarity score of 0.40 mean?
- How face search works, step by step
If you want to see an unnamed identity for yourself, run a face search and watch for a high-scoring result with no profile behind it. Detection happens in your browser; only the cropped face region is sent for matching.