INSIDE THE WORKBENCH

How results are made

MythicNameGenerator uses small curated datasets and fixed composition rules. Generation runs in your browser without live AI, accounts, or a database.

Names are assembled, not translated

Dragon names combine elemental roots, temperament endings and optional epithets. Monsters use separate creature pools. Pirates use given names, roles and epithets. Zombies use human-name components and undead nicknames. Warriors use role-specific given names, tone-based compound bynames and descriptive epithets. AI helped draft the original creative pools and explanations; it is not a source for factual meanings. The warrior pool also includes eight normalized given-name forms selected from the 1897 Onomasticon Anglo-Saxonicum by W. G. Searle. Their role assignments and complete generated names remain fictional; the warrior tool page links to the edition. Fairies, demons, guilds, kingdoms and fantasy cities use their own invented parts, chosen by filters such as court, domain, group type, tone and settlement size.

Japanese names are real; the pairings are not

The Japanese tool draws on published lists: 100 common family names with their kanji, and given names selected from Wikipedia’s lists of Japanese given names. It pairs them at random, so a full name is fictional and describes nobody in particular. Long-vowel marks are removed for easy typing. Meanings are not shown, because the same spoken name can be written with many kanji and we have not sourced them.

Objects are curated; states are factual

The 240 physical objects have eight categories and three suggested drawing-difficulty tags. These judgments are subjective. The state dataset contains the 50 US states, grouped using the US Census regions reference. DC and territories are excluded.

What random means here

We use the browser’s ordinary pseudorandom source to sample an eligible pool. Unique mode shuffles and takes a bounded number of distinct results. Repeats mode samples with replacement. This is creative and casual selection, without cryptographic or exact statistical guarantees. A later batch can repeat an earlier result.

Review and corrections

Automated checks cover data shape, duplicates, filter membership and batch limits. AI review looks for obvious awkward combinations. It is not human approval: the initial human sample and content review remains pending before launch. No tool checks trademarks, character-name collisions or availability.

Report problems through the contact page. No competitor datasets or prose were copied.