Overview: The Plan

I had the idea to make computer generated erotic literature some time ago. At the time of writing, AI generated stories are generally still weak in terms of plot, consistency, and overall cohesion. The output writings are often rambling at best or completely non-sensical at worst.

That said, there is one genre it felt particularly suited for: porn!

My reasoning for this:
– The plot of erotic novels is often similar between novels.
– Erotic literature is often not taken as seriously as other genres.
– Porn has a large audience of potential readers for testing!

From that, we can largely cancel out the disadvantages of AI generated erotic novels. However, there is one massive advantage of computer generated porn: customization!

There are a large variety of body type, scenario, kink, gender, and many more preferences! With AI, the user can set their exact preferences.

At the very minimum, a find and replace of sorts can already give us significant customization and a base story can be made to fit many audiences! The nouns may change (pizza boy, sister, babysitter), the attributes may change (tall, dark, handsome), and the genders may change, but the interactions are as old as time. Computers are relatively good at making these substitutions.

Imagine a mad lib of a story, where your exact desires are used to fill in the blanks! Customize your characters, choose their scenarios, and even choose how you’d like them to interact! It’s all easily within reach though laborious to get right.

My plan for getting there is this:
1.) Markov Chains – essentially have a computer read a bunch of porn, notice what word pairs are often next to each other, and have the computer write a few stories from a base starting point.
+ Fairly easy with lots of libraries and examples
– Often disjointed sentences with little plot, hard to customize

2.) Mad Libs – create erotic novel mad libs. Essentially, create story outlines with blanks for the computer to substitute in a few phrases according to user customization. This will start with character attributes and expand into interactions.
+ Fairly easy, able to start simple, reach a functional prototype, and grow in complexity over time
– Significant amount of busy work to write stories with substitutable phrases, generating all the necessary inputs for customization, making logic descriptive phrases, and eventually covering a wide variety of interactions.

3.) A Hybrid – have a Markov chain substitute in the scenarios between the phrases generated in the previous step. This may not yield results, but I am curious to try it.
+ Building on previous work
– To my knowledge, libraries do not allow this behavior, so it may require significant customization or writing my own library from scratch.

4.) Full Generation – still an unknown at this point. In theory, there may be a way to combine a few generators together. An overall story generator could request weather from a weather generator, a plot point from a plot generator, character description from a character generator, etc. until the entire story is written. Both the story and plot generators are still a fairly big unknowns to tackle and key to pulling this off.
+ Fully customized erotic novels
– Significant research, design challenges, and implementation required. Unknown territory