Creaturator

Creaturator is a little server that vends an API to generate random names and descriptions for legendary creatures. It uses the legendata dataset as input by default.

API

The API currently takes a number (between 1 and 100) of creatures to generate as its only parameter:

https://creaturator.deno.dev/api/v1/{number}

It returns an array of objects, each with name and description string properties. Here is an example of fetching and logging 10 creatures:

const response = await fetch("https://creaturator.deno.dev/api/v1/10");
const creatures = await response.json();

for (const { name, description } of creatures) {
  console.log(`${name} - ${description}`);
}

Input

The server accepts a command-line argument to use as the base URL for fetching data:

deno run --allow-net server.ts --data=http://localhost:8000

It is expected that the following newline-delimited text files can be fetched from the data URL:

1. adjectives.txt
2. attributes.txt
3. item_attributes.txt
4. names.txt
5. species.txt
6. verbs.txt

If no data URL is provided, the legendata dataset is used as the default input.

Implementation

Names are generated using the namerator module, and descriptions are generated by randomly picking parts of speech/phrases to fill in with data. The project is built using Deno, and the server is implemented using Oak.

Future directions

Description generation is currently quite basic, and there are countless ways descriptions could be more interesting. A couple ideas for manual improvements:

1. The description generator "flips a coin" to include adjectives, attributes, and verbs. This could be expanded to optionally include multiples of a part of speech, increase the probability of including more content based on the current length of the description, etc.

2. Some combinations of parts of speech currently don't make much sense, e.g. the API can generate a "flightless" creature that "flies". Metadata/tags could be used to ensure opposing concepts aren't put together (or related concepts have a higher likelihood of being put together). It would probably make sense for this to be done using semantic analysis libraries instead of tracking things by hand.