Story Bot

Experience the magic of AI-generated stories – where bots submit daily stories

About

Story Bot is a blog powered by AI. It receives a prompt and generates a story with a title and an image. Almost all the content on the website has been generated exclusively by AI.

It currently runs with GPT-3.5 but will use GPT-4 as soon as it is available.

Motivation

With the upcoming evolution of AI, I wanted to do a fun project where everyone can participate.

I ended up with this platform, I believe that a blog is something that can survive time and can show us how technology has been evolving. It could be a type of “archive”.

How to submit a prompt

It is super easy! Just open an issue and select the option to write a prompt. It will generate a template to write a GitHub Issue. After this, the system automates the generation of the content.

If you are keen to know more, look into the Prompt Automation section.

Development

Story Bot has three important elements:

• The Client.
• The Story Generation Script.
• A set of GitHub actions to allow users to create prompts.

Before using the client or the scripts, you need to install the dependencies with npm install. The project uses Node 18.

Client

The client was built with Eleventy. The reason for this framework is that the site itself will always be static (there is no need to render anything client side) and 11ty provides great tools for that. It is fast, it brings plenty of plugins, and it can handle Markdown (with syntax highlighting).

So, taking this in consideration, I decided to build the static site on every new prompt. It is fast, uses very little CI resources and doesn’t require a server, just a CDN.

I even took the effort to add support for RSS so that it can even be read from RSS readers and offline systems.

The website is rebuilt after every new commit.

How to run the application

• npm install
• npm start: Will serve a development site.
• npm run build: Will build the site in /dist.

Story Generation Script

The story generation script is a set of tools available in the scripts/ directory.

When you run npm write it will read the contents in prompt.txt and make a request to OpenAI asking for a story, after it gets the story it will request a title for the story and a prompt to generate an image for the story.

Once it has all this data, it uses the image prompt to request to OpenAI’s DALL-E and image.

sequenceDiagram
    participant DALL-E
    actor User
    participant GPT
    User --> GPT: You are an AI that writes blog posts
    User -) GPT: Create a post with prompt.txt
    GPT -) User: STORY
    User -) GPT: How would you title the post?
    GPT -) User: STORY TITLE
    User -) GPT: Create a prompt to create an image for this story
    GPT -) User: IMAGE PROMPT
    User -) DALL-E: IMAGE PROMPT
    DALL-E -) User: IMAGE.PNG

Once it received all the information, it will use 11ty Image to download the image and compress it (if not, the repo will increment it’s size quite fast).

Finally, it will merge all this information into a Markdown file and will be ready to be rendered in the client.

Prompt Automation

The automation process is a set of GitHub actions that runs on every new issue created.

An issue template was created to provide a form for the generation of new prompts.

When an issue with certain parameters is created, the automation runs, and it parses the issue body to obtain the prompt, which is dumped into the prompt.txt file (the file that is used when running the story generation script).

Subsequently, the system runs the script to write the new story, commits those files and generates a pull request.

stateDiagram-v2
    PromptFile: prompt.txt
    Prompt
    Markdown: Story Post
    Issue: Issue with Prompt

    state Writes {
        Install: npm ci
        Write: Runs story generation script
        [*] --> Checkout
        Checkout --> Install
        Install --> Write
        PromptFile --> Write
        Issue --> Write: Author's @ and avatar
        Write --> Markdown: Generate files
    }
    state PullRequest {
        PR: Pull Request
        Commit: Git Commit
        Push: Push new branch
        Markdown --> Commit
        Issue --> Commit: Author's @ and avatar
        Commit --> Push
        Push --> PR
    }
    state New-Issue {
        [*] --> Issue
        Issue --> Parser
        Parser --> Prompt: Parses Issue text
        Prompt --> PromptFile: Dumps content in file
    }

The reason for which it creates a Pull Request instead of committing directly is so that a human factor can verify that the content is appropriate to the blog. We don’t want anyone jeopardizing this blog by adding unethical prompts.