A JAV code splits into four layers: a label prefix, a serial number, an optional internal padding used by retailers, and optional suffixes bolted on downstream. Only the first two come from the studio. Reading a code correctly means knowing which layer each character belongs to, and discarding the layers that were added by someone else.
Take 118abf00042-4K as a worst case. The studio issued ABF-042. Everything
else is packaging: 118 is a retailer's maker prefix, the padding to five
digits is a database convention, and -4K is a claim made by whoever renamed
the file.
What are the four layers of a code?
Each layer is added by a different party, at a different stage.
| Layer | Example fragment | Added by | Reliable? |
|---|---|---|---|
| Label prefix | ABF |
The studio | Yes — identifies the label |
| Serial number | 042 |
The studio | Yes — unique within the prefix |
| Internal padding / maker prefix | 118…00042 |
The retailer's database | Yes, but not for searching |
| Suffix | -4K, -C, -U |
Resellers, subtitle groups, uploaders | No — a claim, not a fact |
The practical rule that follows: search layer 1 + layer 2 only. Strip everything else before you type. If you need a specific version, add the suffix back afterwards as a separate refinement.
How does the prefix work?
The prefix is a short alphabetic tag owned by a label, and labels retire and replace prefixes over time.
Prefixes are typically two to five letters. They are not registered anywhere centrally, which is why collisions and near-collisions exist across small studios. What matters in practice is that the big labels rotate them: when a line is refreshed, rebranded or moved to a new production standard, the label opens a new prefix and lets the old one stop.
Three well-known rotations:
| Older prefixes | Current prefix | Label | Changeover |
|---|---|---|---|
SSNI → SSIS → SONE |
SNOS |
S1 No.1 Style | SONE from Dec 2023, SNOS from Nov 2025 |
IPZ → IPX |
IPZZ |
IdeaPocket | IPZZ from Mar 2023 |
ABP → ABW |
ABF |
PRESTIGE | ABF from Jul 2023 |
Changeover months taken from the first and last serials listed in each label's own
catalogue and in FANZA product records, checked 2026-08-03. Note how tightly the
handovers run: SSIS closed at 997 in February 2024, two months after SONE-001
shipped, and SONE closed at 998 in November 2025, days after SNOS-001.
A retired prefix does not become invalid — the back catalogue keeps its numbers forever. It simply stops receiving new ones. That distinction matters when a search returns nothing: see codes from studios that no longer release.
How does the serial number work?
The serial is sequential within a prefix, starting at 001, and it orders releases only approximately.
Numbers are allocated in advance of release, often in blocks, and a label may put four or five titles out on the same street date. Numbers also get skipped — a cancelled project leaves a gap that is never filled. So a serial is a reliable identifier and an unreliable timestamp. If you need a date, take it from a catalogue entry, not from arithmetic on the number.
Serials generally run to three or four digits before a prefix is retired, though long-running planning series push much higher.
Why do retailers show a different, longer ID?
Because storefront databases normalise codes into a fixed-width internal form, and some prepend a numeric maker identifier.
The transformation is mechanical:
- Lowercase the prefix —
ABFbecomesabf - Pad the serial to five digits —
042becomes00042 - Optionally prepend a numeric maker code —
118abf00042 - Optionally prepend
h_for certain catalogue segments —h_1236abf00042
The numeric part tracks the distributor, not the label. Reading FANZA's own
product IDs, every Prestige title carries 118 (118abf00042, 118abw00001,
118docp00100), everything moving through SOD's distribution carries 1
(1stars00001, 1rctd00100, 1fsdss00001), Alice Japan carries 53, Max-A
60, Glory Quest 13. Several of the largest labels carry no numeric prefix at
all — S1, MOODYZ, IdeaPocket and Madonna appear as plain sone00001, midv900,
ipzz00001, jur00001 (FANZA product records, checked 2026-08-03).
That is why the digits are worth recognising rather than memorising: the same
letters under a different number are a different maker. 436abf00042 and
118abf00042 are two unrelated releases from two unrelated companies.
All of these resolve to the same release. The reason to recognise the pattern is
that these internal IDs leak into file names and third-party listings, and they
look like corrupted codes if you have not seen them before. When one appears,
mentally reverse the steps: strip leading digits and h_, strip leading zeros
from the serial, reinsert the hyphen. 1sdab00123 is SDAB-123.
How do you read the optional parts?
Everything after the serial belongs to one of three categories, and they mean very different things.
| Fragment | Category | Meaning |
|---|---|---|
-A, -B, -CD1 |
File splitting | One release cut into multiple files |
-C, -CH, -SUB |
Language | A subtitle track was added by a third party |
-U, -UC |
Processing claim | Claimed uncensored or de-censored version |
-4K, -FHD, -HD |
Quality claim | Claimed resolution of that particular file |
hhd800.com@, leading site tags |
Distribution noise | Added by a mirror or aggregator |
Only the first category tells you something structural about the release. The rest describe a file, not a title, and none of them are verifiable from the name. Suffixes get their own breakdown in what do the suffixes -C, -U and -4K mean?.
Site tags prepended to the code are the most disruptive kind, because they break alphabetical sorting and defeat exact-match search. Strip anything containing a domain before searching.
What does a code look like when it is not a studio release?
Non-studio releases follow entirely different shapes, and recognising the shape tells you which catalogue to look in.
| Shape | Family | How to read it |
|---|---|---|
MIDV-123 |
Studio label | Prefix + serial |
259LUXU-1234 |
Amateur / planning series | Numeric series ID + label + serial |
FC2-PPV-1234567 |
Individual seller platform | Platform tag + global serial |
010124-001 |
Uncensored subscription site | Release date (MMDDYY) + daily sequence |
HEYZO-1234 |
Uncensored subscription site | Site name + sequential serial |
The date-shaped ones carry information the studio format does not: 010124-001
tells you the release date exactly, because that is the release date. What it
does not tell you is the site, which is why the same-shaped code from two
different sites is genuinely ambiguous without context.
What if the code is right but nothing comes back?
At that point the code has stopped being the useful handle, and the image is.
A correctly typed, correctly stripped code that returns nothing usually means the release is outside the catalogue you are searching — not that the code is wrong. When that happens, identifying the performer from a frame is often faster than continuing to chase the code. Our index holds 241,792 faces matched by 512-dimension face vectors (our index, 2026-08 snapshot), with a similarity threshold of 0.40 for video stills.
Boundary worth stating: that index is built from faces, not catalogue records. It has no release dates, no maker fields and no per-prefix inventory, so it can tell you who is on screen and cannot tell you which code you are holding.