- Amalgam: This is the default mode as it generates the catchiest results. It uses a mixture of words that sound cool to begin with (e.g., existing brands, artists, superheroes, and even cocktails). No custom input is required.
- Dictionary: This mode generates words based on a large dictionary, comprising the 122,800 most common words on the internet, most of which are English. No custom input is required.
- Custom input: This mode requires you to write or paste your own words into the text box, up to a total of 4096 characters. You may separate your input words however you want, as long as you do not use
/!";'<>or any other forbidden characters. - Min. and max. length: Limits the length of the generated words to a specified range. Whether the maximum length is reached depends on the mode and your custom input.
I named this little project “Brand Name Generator” because I figured it would be closest to what you might search on Google. However, it can be used to generate all kinds of artificial words for various purposes, especially online pseudonyms (which, admittedly, might be considered brands as well). Feel free to get creative with the results; they are intended to inspire, not to be taken literally.
I conceived the idea for the generator when my friend Lars-Ole Kremer was looking to name an electronic music project. I am still unsure what it was—something between a pseudonym or DAW pipeline or even plugin—but I came up with “Fabletron,” an amalgamation of the words “fable” and “Ableton.” We added the additional “R” due to fewer existing results on Google. Essentially, this is all the generator does: combining two words, taking the first half from one and the second half from the other.
Because hyphenation rules are quite complex and their implementation would not even improve the algorithm that much, I used a simple approach to tokenize the input words, which had proven itself during my projects Babaism and Fantasy Name Generator: words are split up wherever a vowel meets a consonant. This yields consecutive groups of characters, such as p, thr, ai, or you (for simplicity, y is always counted as a vowel). This allowed me to express words in an ABABAB... form, A being vowel tokens and B being consonant tokens. The tokens are learned from both the beginning and the end of a word, up to a certain length, yielding seven different expressions in total:
Example: “custom” and “input”Groups: AB, BA, ABA, BAB, ABAB, BABA, BABAB1st halves: in/inp, cu, inpu, cus/cust, input, custo, custom2nd halves: om/ut, –, –, tom/stom/put/nput, ustom/input, custom
The combinations are then chosen at random, which could lead to outputs as diverse as intom or cusinput. Naturally, the number of unique permutations is limited, so using a larger input word base greatly increases the variety of output words.
Some combinations were impossible to read, let alone pronounce, which had to do with the way they were combined; for example, hallchriever or epistrclutch, which would be contracted or reduced somehow under normal linguistic circumstances. Therefore, I decided to discard words that are welded together by a letter combination found nowhere in the dictionary. For this, as well as for a large base of input words, I started with my favorite collection of the 122,800 most common words on the internet, introducing the “Dictionary” mode.
Because cycling through this many words repeatedly, checking whether an output word contains certain syllables, is extremely inefficient, I chopped up the entire dictionary into unique syllables of types AA, BB, AB, or BA and stored them as a binary file to decrease loading times. In fact, I managed to optimize the hell out of the algorithm, increasing performance nearly hundredfold by pre-building binary files and using Python’s set() datatype instead of list() wherever possible. Without going into detail, sets contain each value just once and without any order. Their main advantage is that items are stored by their hash value, which makes searching for specific items very efficient.
Being able to generate artificial words never seen before by anyone was a huge step. Nonetheless, the 122,800-word dictionary lacked one essential feature: semantics. Frankly, it was impossible to use the algorithm for a specific task, such as creating words only from chemistry or music vocabulary. Thus, I began making my own input word collections using ChatGPT 4o. I would always prompt something like, “List 300 fashion labels, separated by commas, absolutely no duplicates!” As of today, ChatGPT is still unable to admit that it cannot find that many unique words, always producing lots of duplicates, often triplicates or quadruplicates. But it did not matter as those are removed by the set() feature.
On the other hand, ChatGPT excelled at finding interesting categories that contain cool and catchy words by default. So far, I have included about 300 examples each for the following categories: brands and corporations, products, apps, domains, tags and slogans, band names, songs, albums, fictional places, books, magazines, comics, superheroes and villains, common nicknames, famous mascots, sports teams, conventions and festivals, artwork and paintings, colors, cars, electronics, fashion lines and collections, podcasts, TV shows, movies, video games, board and parlor games, fancy English words and cocktails.
Just about now, I decided how boring it would be to generate new fashion labels only based on existing fashion labels. Hence, I put all words together again, forming the “Amalgam” mode, which is the default mode you can try out at the top of this page. In my opinion, the results look less awkward than those for the Dictionary mode, but results might contain the occasional Netflixxx or Metallicar. I am no expert in corporate or copyright law, but if you want to be safe, do not use anything that looks similar to an existing brand when building your own. In fact, another idea I am toying with is including a feature that indicates how popular the created words are on Google. This could be used to rule out existing brands automatically. However, this exceeds my current programming skills. Consequently, you have to Google your favorites yourself and decide whether a SoundCloud profile named Dazebub or Explodyssey with only six followers is reason enough to name your brand differently.
To get back to the idea of semantic word categories, I incorporated some early user feedback and introduced a third mode, the “Custom Input.” This at least allows you to write (or better, paste) your own input words, generating artificial words independent from my Amalgam and Dictionary word bases. Your input is processed with a high degree of flexibility, accepting all kinds of delimiters (word separators) and removing any non-alphabetic characters (symbols and numbers). As this required an additional textbox on the webform, I had to do some PHP bombproofing to prevent malicious inputs, potentially XSS attacks. If you are a white-hat hacker, please tell me how poorly I did. If you are a black-hat hacker, please go hack somebody else.
The Custom Input feature gave me a hard time during debugging as the algorithm was not very robust, often getting stuck in death loops when the custom input text was sufficiently short. It all came down to the algorithm expecting to produce a given number of output words, which could never be reached due to the limited number of permutations. Two stopping criteria later—a number of unsuccessful attempts as well as a time limit—and the Brand Name Generator was good to go. Alas, I also learned the hard way that developing scripts for a web server on a web server should be avoided as much as possible, especially when only having a mobile hotspot at hand. Do yourself a favor and bypass the online components when debugging offline features.
Last but not least, let me acknowledge my talented sister Marlene (aka NannoCloud), who painted two-thirds of the banner image for this page. I will let you guess which parts. Nonetheless, be sure to check out her Instagram and homepage!
Leave a Reply