zestor/muses

Artificial Intelligence (AI) Long Form Fiction Writer using OpenAI or NLP Cloud

Muses

The intent of this program is to produce long open-ended fiction using cloud based AI engines like OpenAI and NLP Cloud Playground.

The code expects your tokens to be stored in environment variables as OPENAI_API_KEY or NLPCLOUD_API_KEY, or both.

start the program with % python muses.py

1. The program has been primed with a previous_scene, characters, summary, and protagonist. The main loop writes 14 paragraphs to avoid writing forever, but the loop could be increased or decreased.

   • previous scene - what happened last. If you are going to restart the program to pickup where it left off, start with a short previous scene summary instead of the long text produced as an AI response.
   • characters - you can add a name and short description of your character. The program will continue to add to this description as the story continues and things happen to your characters.
   • summary - an ongoing summarization of what has happened.
   • protagonist - a name of a character who cannot die. Prompt is written to not kill or injure this character.

2. The output will be written to the file story.txt. It will also clear this file on every restart.

3. The generated scenes will be written to SceneOutput. It will also clear this directory/folder on every restart.

4. The program will insert randomness to drive the story in different directions using a plot queue. These are simple little text strings that are something like: someone in this story is battling with a past mistake. This helps the program to generate an open-ended ongoing story with some variety from scene to scene.

5. Main Loop

• Create a new scene using the previous scene, characters, plot, summary, and protagonist data
• This program supports parallelism, and can call the AI cloud engine simulaneously multiple times for the same scene generation. The intent is to pick the best of X scene generations. Best is determined by text length. Meaning the longest text length will be selected. Default is best of 3 attempts. Parallel calls work fine for openAI, but you will want to scale the parallelism down for NLP Cloud to probably 1 to avoid errors depending on your engine used and paid plan.
• once the longest scene text is generated, it is sent again to the cloud AI engine to enhance the scene description with vivid sensory details. This increases the text length with further sensory details while keeping within the original intent of the scene.
• there is logic to avoid duplicate output or repeating verbatim sentences in future scenes if it's already been written.
• Next the program determines if any new characters have been added to the story by the AI engine
• Character details are updated based on what just happened in this scene, and any duplicate character descriptions are summarized to the character and any duplication removed.
• The summary is updated with what just happened in this scene.

muses's Issues

The Install Not Working

I followed these instructions in muses.py

Created
~/.bash-profile
Added the following two keys
export OPENAI_API_KEY=
export NLPCLOUD_API_KEY=
Saved the file and sourced it.
source ~/.bash-profile_

So far so good. However, when I run the app -
python muses.py

This is the error:

Traceback (most recent call last):
File "/home/robert/Muses/muses.py", line 7, in
from python.MusesHelper import MusesHelper
File "/home/robert/Muses/python/MusesHelper.py", line 9, in
import cohere
ModuleNotFoundError: No module named 'cohere'

So thank you for creating this, but without some assistance I'm unable to use this.