Band names that fit the genre

A generator that hands you a name and says nothing else is asking you to take it on faith. A band name is the wrong thing to take on faith: if you name a band, print a hundred shirts and then discover the name belongs to someone else, you have a trademark problem rather than an awkward conversation. So every name is checked, in your browser, before you see it.

Four tiers, in descending confidence

An exact match is the same string. A normalised match is the same name once case, punctuation, diacritics, an ampersand and a leading “The” are set aside — so a name and that name without its definite article are correctly treated as one name, and an initialism written with full stops matches the same initialism written without them. A variant is the same name with a plural or a spelling change. A near match is a name one or two characters away, a name containing a real act's whole name, or a name built on a word that essentially only one real act uses.

Why the near tier is deliberately hard to trigger

An over-strict checker is worse than no checker. If a rule flags every two-word name that contains an ordinary noun, you learn within one batch that the warnings are noise, you start skipping them, and then you skip the one that was real. So a shared word only counts as evidence if it is genuinely distinctive — a coinage or a proper noun, not ordinary English. That distinction is drawn using an external dictionary rather than this app's own corpus, because a corpus this size makes perfectly ordinary words look rare, and tuning against it would have measured the same skew twice.

What it actually catches, measured

Two corpora of real acts and two sets of invented names were commissioned from independent agents who never saw this code. One pair was used for tuning; the other was scored once, at the end, and is reported here.

On the held-out set of real acts, the check flags 54.4% overall — but that single number hides everything useful, because it depends entirely on how well known the act is. It catches 100% of the most famous acts, 80% of very well known ones, 62% of acts known to people who follow music, and 38% of genuinely obscure ones. On the held-out set of invented names it raised a false alarm on 0.7% — three names out of 418, every one of them the softer “close to” warning rather than a claim that the name was taken.

This is a corpus-membership test, not an originality test. A clear result means the name is not in our list. It does not mean the name is available. The list holds 1,775 well-known acts; there are hundreds of thousands of bands, and the list is Latin-script only. Before you commit to a name, search a trademark register and search a streaming service. Those two searches take a minute and this check does not replace them.

Where the misses concentrate, and why

The false negatives are not spread evenly, and the reason is structural. While the bundled examples were being written, the writer kept reaching for a name and then recognising it as an act that already exists — a funerary abstraction, a small-society name, an honorific followed by a short name. Eighteen confirmed acts found that way are now in the corpus.

The lesson generalises. The devices are not too weak; they are too well aimed. A naming device encodes what a scene's names actually do, so a good device points at the same small region of name-space the scene's real bands already occupy — often at the very band that established the convention. The better the device, the likelier its best output is taken. Misses therefore concentrate exactly where this app is working well: plausible, idiomatic, scene-correct names belonging to acts too small for a list of this size. That is the argument for searching properly especially when a name comes back clear and sounds perfect.

Twenty-two genre families are modelled as inventories of naming devices — rules of formation, each tagged with the semantic field it draws on and the structural shape it produces. A metal name and a twee-pop name differ structurally, not just in vocabulary: one is frequently not in English and rarely takes a definite article, the other reaches deliberately for the small and the domestic, post-rock names are often complete sentences, and hip-hop names are identity-forward and place-coded.

There are no example band names anywhere in the prompt, deliberately: a sample name is not an illustration, it is an instruction, and its shape comes back in every output. The inventory is handed over as a vocabulary for describing finished work, never as a checklist to work through.

Batch sizes are chosen so they cannot evenly divide the device inventory. Every inventory holds more devices than the largest batch you can ask for, so covering the set is never the cheapest way to look varied — “use each one once” is impossible by arithmetic rather than by instruction. If you type a genre we do not model as its own family, the panel above the run button says so and the generic devices are used instead.

Variety is then measured where it actually hides. Batch openings are compared two content words deep — because “making a promise” and “making the promises” are the same opening — alongside the spread of structural frame and semantic field across the batch. A rarity-gated near match works the same way: a shared word only counts as evidence when its rarity is real.

Every name comes with the naming convention it uses, why that convention fits the genre, and how the name behaves as a wordmark — whether it survives being set small on a shirt or turned sideways on a spine.

The model also gives its own read on whether a name is likely taken. That read is reconciled against the browser corpus check, and any disagreements between the two are surfaced as findings rather than quietly resolved — the model guessing “probably fine” about a name our list already holds is exactly the thing worth seeing.

Honest limits are stated in the interface, not just here. Every result carries the same caveat this page does: a clear verdict means the name is not in our list, never that it is available, with links out to a trademark register and a streaming search.

A finished batch has a one-click export as Markdown, JSON or CSV, including the corpus verdict for every name, so the shortlist you take away carries its own evidence. Signed in, shortlists save to your account as threads, so asking for more names continues the same search instead of starting a new one.

Cost

Naming runs on a metered model and needs an account. The figure beside the run button is a hold, not a price: it reserves against the full output cap, and what is actually charged is usually well below it. The minimum shown beside it is the floor a run needs to start at all.

Everything else on this page is free and needs no sign-in: the genre convention readout, the corpus check, the standalone name checker, the variety measurement, and the three saved example runs. Those cost nothing because they never leave your browser.