A failed search almost always has one of four causes, and they need different fixes. Diagnosing which one you are hitting takes about a minute and saves you from repeating the same query with slightly different spellings.
The four are: weak input, unnamed identity, outside the index, and wrong name form. The symptoms differ enough that you can usually tell them apart from the results screen alone.
How do I tell which failure I'm hitting?
Read the results, not just the absence of the answer you wanted.
| What you see | Most likely cause | What to do |
|---|---|---|
| "No face detected" | Weak input | Pick a different frame; see the screenshot guide |
| Results returned, all below 0.40 | Weak input or outside the index | Retry with a second frame first |
| Strong score, but the result has no name | Unnamed identity in the index | Use the returned reference images and similar-face lists |
| Correct performer found, but not this title | Title outside the index, or different name form | Search by code or by performer filmography |
| Code search returns nothing | Wrong name form, or formatting error | Check separator, prefix and suffix conventions |
The second row is the one people misread most often. A list of results is not evidence that the answer is present — vector search always returns its nearest candidates, even when the right person was never indexed.
Why does a strong match come back without a name?
Because indexing a face and identifying a face are separate steps, and the second one does not happen automatically.
Our index holds 241,792 faces, but only 2,333 are linked to a named performer record (our index, 2026-08 snapshot). The gap is not a bug — most indexed faces come from covers and stills where no reliable name was attached to the image. Our similar-performer data alone contains 7,004 references pointing at identities with no name, spread across the similar-face lists of 1,899 named performers.
Practically: if you get a high-scoring unnamed result, you have found the right face. You now need a different route to the name — the reference image it came from, the site it was sourced from, or the named performers that identity sits closest to.
Why would a title be missing from the index entirely?
Because no index covers everything, and ours has a shape you should know about.
We index faces from 106 sites. But 2,246 of our 2,333 named performers have their representative image sourced from a single site. That is a heavy concentration, and it defines the blind spots:
- Releases that circulated only on catalogues we do not cover
- Older material that predates the sources we index
- Amateur and individually-produced content, which is rarely aggregated with consistent metadata
- Titles removed from every source we index before we saw them
None of these produce an error message. They produce a plausible-looking list of the wrong people, which is why reading scores matters more than reading rank order.
Why does the same title have several names?
Because a release accumulates identifiers as it moves between catalogues, and searching one form misses the others.
| Form | Example shape | Where it appears |
|---|---|---|
| Studio code | Letters + digits, often hyphenated | Cover art, official catalogues |
| Alternate code | Same letters, different separator or padding | Aggregators, databases |
| Japanese title | Original release title | Official listings |
| Translated title | English or machine-translated | International aggregators |
| Suffixed variant | Code plus a trailing marker | Re-releases, format variants |
Translation is a large and under-appreciated source of misses. In our own performer data, 197 records have a name whose English rendering differs from the primary name form — meaning the same person is findable under two spellings depending on which source you started from. The same divergence happens to titles, at greater scale.
What is the most efficient order to try things?
Escalate from cheapest to most expensive:
- A second frame. Free, takes ten seconds, and resolves a large share of failures caused by pose or blur.
- A different person in the scene. Group scenes give you multiple shots at identification.
- A code, if one is visible. Codes are burned into intros, covers and watermarks more often than people notice. A code converts an image problem into a text problem.
- The performer's filmography. Once you have a name, narrow by scene details — wardrobe, hair length, setting, on-screen text.
- Accept absence, provisionally. Some material is genuinely not in any index you can reach. Re-checking in a few months is more productive than re-phrasing the same query today.
A note on honesty: we would rather tell you the index does not contain something than show you a confident wrong answer. That is also why scores are shown on every result. A tool that always finds something is not better — it is just quieter about failing.
Related questions
- How to find a video from a single screenshot
- How to find a video by its JAV code
- How to take a screenshot that face search can actually use
- The video was taken down — how to find it elsewhere
- Common mistakes people make when searching for a source
If you have not yet tried a second frame, that is the highest-yield next step. Upload it here — detection runs in your browser, the original never leaves your device, and the score on each result tells you whether it means anything.