As an experienced author, I’ve been working on some ways to work with GPT3 in creative projects.
I’ve built a github repo with some of the tools I’ve been developing. I’d really appreciate feedback. It’s very much a work in progress…
I’ll post a youtube video next week (once I get some credit back on my openai account!) but in the meantime, please go and check this out and let me know what you produce!
In brief, the tools I’ve been experimenting with:
1- create and develop character outlines based on whatever inspiration you have.
2- create story synopses based on those characters - or based on any other inspiration you want to give - i.e. description of genres ideas for settings philosophical conjectures poetry news stories …whatever.
3- break those synopses down into scene outlines - and then break those scene outlines down even further to get more and more detailed story plans.
4- at any stage, you’re working with text files, so you can break in and make changes, introducing your own ideas and edits to improve the story - This is a vital part of the process. I’m not trying to create an automatic system. You’re still the writer, so you need to do some creative work!
5- finally, there’s a file here which will turn your scene outlines into film scripts. I kid you not.
Note: Results depend very much on the quality and detail of your scene outlines. You’ll have to do a lot of work to tidy up and improve your scene outlines to get a good script.
Thanks particularly to @daveshapautomator for his great work on this topic (and I’ve used some of his code here)
This is pretty cool. You might consider including some example outputs from each script in your README. That will make it easier for folks to know what to expect.
I haven’t trained any new models. Was working on creating a data set with a bunch of papers on consciousness in AI, DMT, and LLMs.
Almost all of the papers are PDFs with multiple columns. The simple cloud pdf to txt sites read contents from both columns as one line. Next I found the Adobe API, which seems very powerful. There I ran into a weird module error, and have not been able to translate the pdfs into useable txt yet.
I really want to get the pdf to txt pipeline working because I think that will be one of the richest sources of material. What I have used so far has been off Gutenberg, which has limited options.
Not sure what the best way of training would be, I’m thinking of going very micro with one word prompt:response, will start a new thread on that.
Have spent a few nights manually generating creative writing content to guide an eventual automation process. Very excited to see some tools already in place!
When I’m “freestyling” on a small amount of content, I find that text002 creates responses that are a bit too “constrained”. So if I want to add some creative language dimension to a paragraph, I’ll use one of the fine-tunned models.
Definitely like the idea of creating a movie script as I find that text002 does really well when given a lot of content and context to work with.
One thing I’m trying to keep in mind is that compute resources are precious. I see no point in generating something that I myself would not care to read. For that reason, I like to be closely guiding the model, and updating my own neural network on the optimal ways to interact with the model.
I trained GPT3 on few classical books.
One of them is “hitchhiker’s guide to the galaxy”.
It generates funny text sometimes. But it’s more like idea generator.
So you click 100 times for random text and pick up manually some funny ideas.
It definitely doesn’t suite for write the whole story.
I just took the whole book.
And split it as 1 sentence as an input and 3 followings sentences as an output.
Stupidly simple and maybe not effective at some point )
sir, I am getting the following error in 3Break down synopsis or scene into list.py
File “c:\Users\hanle\git\Creative-Writers-Toolkit\3Break down synopsis or scene into list.py”, line 112, in
scriptprompt= prompt8.replace(’<>’, completion7).replace(’<>’,breakdownexample).replace(’<>’,extrainfo).replace(’<>’,numberofscenes)
NameError: name ‘extrainfo’ is not defined
by the way, the version I’m working on now replaces SCENE001.TXT files with SCENE0001.TXT files (so you can work on projects with more than 100 scenes). This means you’ll have to rename your SCENE files with an extra leading zero when I put the new version up. sorry.
I think you mean line 88, which already has that line of code in it. Maybe a “try except” statement might fix it…I tried one but didn’t get that to work.
My story is called “The Great Slowdown” and is about collapsing supply chains and contraction of the energy supply. It is set in the suburbs and follows a family’s journey. There are also elements of consciousness, aliens, self work, sci-fi, and sustainability/permaculture.
Try putting that line where i suggested - ie before the if statememt. Thar way it makes sure the variable is always defined as a string before its used.
ok - the new version of my writer’s tools is up on github - including a new ability to turn any character into a chatbot - so you can have live conversations with the fictional people you create.
I’ve also uploaded a bit of a tutorial - please, like, subscribe, and comment - I really want others to get involved with this project with ideas and code…
