undefined | Better HN

0 pointssoultrees2y ago0 comments

As someone who has been majorly affected by the Hollywood strikes, I finally dived in and spent the time needed to make mistakes to continually learn how to program.

That said, ive been working on a node.js PDF Text screenplay parser that will ultimately push out a JSON object according to an initiative to standardize all screenplay objects in screenJSON[0].

Here is the github for my project, it’s messy but I’m about 80% there and just need to spend more time on Regex to catch edge cases. Currently working on a next.js front end that maps out the script into a visual format but that’s not ready to be seen yet. Also don’t judge me on the name, it’s not going to stay as that. No AI as of yet, but could use text classifiers to be able to tell the difference between a key prop, and say, a camera movement but I’m trying to get as far as I can without LLM help of the classifications.

https://github.com/mattrichmo/ScreenBaby.

[0] https://screenjson.com.

0 comments

ajani2y ago

Have you tried using any LLM for this task? LLMs are text-in text-out systems and excel at transformations of text.

soultreesOP2y ago

Oh for sure, but since most screenplays are standardized to a point, the idea is to just use Regex to get there, then use a local version of llama 2 to finish up the classifications.

Where the fun with LLMs come in, is after all the screenplays have been parsed and turned into a dataset that is trained on not just the story but also the cinematography aspects, as well since we’ve broken it down to a granular detail of each element of each scene.

ajani2y ago

I see where you're going with this. Elemental (motif-based) editing instead of timeline and frame based editing?

1 more reply

j / k navigate · click thread line to collapse

0 pointssoultrees2y ago0 comments

As someone who has been majorly affected by the Hollywood strikes, I finally dived in and spent the time needed to make mistakes to continually learn how to program.

That said, ive been working on a node.js PDF Text screenplay parser that will ultimately push out a JSON object according to an initiative to standardize all screenplay objects in screenJSON[0].

https://github.com/mattrichmo/ScreenBaby.

[0] https://screenjson.com.

0 comments

ajani2y ago

Have you tried using any LLM for this task? LLMs are text-in text-out systems and excel at transformations of text.

soultreesOP2y ago

Oh for sure, but since most screenplays are standardized to a point, the idea is to just use Regex to get there, then use a local version of llama 2 to finish up the classifications.

ajani2y ago

I see where you're going with this. Elemental (motif-based) editing instead of timeline and frame based editing?

1 more reply

j / k navigate · click thread line to collapse