#Outline Editor Tool for Writers

25 messages · Page 1 of 1 (latest)

plucky jolt
#

Hey guys, I built an outline editor tool for writers. Aim of this tool is to use GPT to assist the writer by suggesting chapters and points that go along with the story that they are currently writing.

Totally free to use, of course. Feel free to check it out - https://pen-pal.vercel.app/

tutorial-gif.gif
rustic escarp
#

Hi. I like your app. But in my mind for working on a novel, you need a global context that applies to the entire text. And the local context (characters, time, weather, place, etc.) is already applied to specific scenes.

#

You can also try using the snowflake method. When each sentence of the plot summary is gradually written down, first into chapters, then into scenes, and then into paragraphs.

sand tulip
#

If you keep a segment of the prompt entirely around how the "story arc" should form together then with each chapter iteration it will know how far it is / what part of the arc has been reached and write accordingly. Just one way to keep story structure

plucky jolt
#

Thanks for the feedback!!

plucky jolt
plucky jolt
rustic escarp
# plucky jolt Do you mind explaining this abit more?

Here are a couple of ideas of what an algorithm for generating artistic text might look like.

  1. Creating a global book context:
    1.1) Genre selection
    1.2) Header selection
    1.3) Choice of text style
    1.4) Brief description of the plot from a person. Within one paragraph 100-500 characters.
    1.5) Generation of a description of the main characters of the book: names, appearance, biography, motives
    1.6) Generation of a description of the world of the book. Geography, historical era, characteristics
    1.7) Generation of a list of minor characters with a brief description of their appearance, biography
    1.8) Generation of a list of book chapters with a brief description of the events of each chapter.
    1.9) Global context generation: text within 1000 characters that summarizes everything to create a context for the GPT-3 neural network.
#

  1. Work on the chapter in order, in order of priority
    2.1) Considering the global context, we generate a detailed description of the events of the current chapter in 1-3 paragraphs (300-500 characters)
    2.2) Create a list of scenes taking place in the chapter. With a brief description of each scene in 1-3 paragraphs, given the global context of the novel.
    2.3) Create a chapter local context within 1000 characters, summing everything up for the GPT-3 neural network context.
    After creating a sequential description of the events in the next chapter and a list of scenes, we move on to step 3, which will include working on each scene, taking into account the global and local context.

  2. Work on the scene in order of priority. The list of scenes for each chapter is given by step 2.2.
    3.1) Identification of the main and secondary characters involved in the scene
    3.2) Description of the venue of the event: time of day, weather, description of the environment.
    3.3) Based on this information, taking into account all the previous text, the next paragraph is generated using GPT-3.
    After completing a scene, start working on the next scene.

plucky jolt
magic jacinth
#

leave letter

deep geyser
#

I am working on a very similar app for screenplays. My problem is having the AI remember and refer to the existing acts and scenes while suggesting new scenes, dialogue, callbacks etc. How are you keeping track of the overall corpus of the story in generating new scenes?

plucky jolt
#

@deep geyser also, cool that you're working on something for screenplays! do you have a link to it?

deep geyser
#

Just a local repo for now. It's too up in the air at the moment to have anything meaningfully shareable. I'm starting by importing an existing script from Final Draft (XML) and turning it into JSON (saving into a mySQL database). I want to add sections for characters and various notes about the story, tone, beatsheet, scene breakdowns, off-the-page story knowledge (i.e. stuff not visible to the audience) and have the AI assist in generating new scenes from the very highest level (i.e. 1 sentence) all the way down to generating character dialogue based on previous scenes. I've had really powerful conversation with ChatGPT but the problem is that by the time I give it enough information to generate something I'm looking for it starts forgetting or making up story information it was already told. ChatGPT was suggesting I use history or context parameters, but it appears that those are deprecated.

plucky jolt
deep geyser
#

No, just davinci API which is pretty robust TBH

#

The interface is in VueJS

plucky jolt
deep geyser
#

It even has its own little AI Chat tab for those times when ChatGPT gives you the cold shoulder but the API is still your friend 😉

Screen_Shot_2023-01-24_at_10.38.10_PM.jpg
plucky jolt
#

niceee

deep geyser
#

And I'm using contenteditable and some keyup events to make it editable.

plucky jolt
#

Added a Global Context feature:

  • Genre Selection: User can select a genre.
  • Plot Summary: User can write a summary of their plot.
  • World Description: User can describe the world in which the story takes place. Geography, historical era, etc.
  • Character Codex: User can describe their characters in detail. Physical appearance, origins, motives, etc.

All the above will be included in the AI prompt whenever the user clicks the 'AI Suggestion' button, which will affect the points it generates. This will add a global context for the story in which the AI can operate. All optional.