I asked it for a story lines process.
The first thing on the list was generating ideas.
Then I asked it for some properties and/or categories for story ideas.
then for each property you could have it give you a list of selections. then ask it for properties to evaluate a storyline idea.
then you can have it be a critic that uses those criteria to evaluate the list of ideas.
then have it try to improve each of the storylines based on the critics advice.
then have the critic reevaluate the storylines based on the improvements.
then move on to the next step and have it go through a similar process.
You can also search for 13B or 70B on huggingface and sort by date to see what others are cooking up.
The kobold API and the kobold.cpp client are also great ways to self host, as they are literally built for storytelling and predate GPT3.5 and Llama.