We are AI technologists and computer scientists, who also run an entertainment production company.
Like many posting here, we also want to make our own GPT-3 movie scripts, but finally a really good one. We’ve figured out a great Prompt + Settings Preset combination for writers to use. You can easily enter parameters and keywords, which helps guide the log line and screenplay idea generation.
This isn’t an “app” but a way to help creative people produce rockin’ content, in writing or someday on TikTok in full VR - wherever, whenever, let’s just make great movies! GPT-3… This is the way. We’re already using it to make a kids science show but we believe with your help we can make Academy Award winners.
Please do not hesitate to contact me, if you’ve used GPT-3 and are interested in coming up with cool TV shows and movies based on it’s suggestions and writing. First just log lines and scripts, sure, but one day the whole enchilada!
CyberFilm AI
I wouldn’t be surprised if Netflix snuck a few AI-produced movies in there, for a few nominations this year Fair game is right
Probably easier streaming, than playing in theaters, for winning any Oscars during the pandemic.
What’s your favorite video you’ve seen so far @m-a.schenk (link please)? I’ve seen a few but nothing good - even from the big players!
Super interesting use-case . Have you tried using finetuned models instead of prompted ones? From my own experiments you can get much much lower latency, as well as higher quality that way. There are also lots of possibilities with optimizing for specific tasks with curated datasets ie tuning the AI-assistant to write spec scripts in the style of A24 horror movies. Also the base models can’t make references to events post 2019 but finetuned models can if they’re trained with the data (ie more fresh and relevant takes).
Also from my own prototypes I’ve found that wrapping GPT3 with a flexible UI can enable lots of interesting AI-assisted workflows ie generating a rough draft of a scene, then have the script writer annotate the draft with comments, then perform a re-write incorporating those comments.
Great idea, not yet. I’m still collecting scripts/metadata but when I have a good set, will try the fine tuning beta. Looks like exactly what I need! Agree some clean script data that is nicely tagged and structured, may give a bit of needed steering, to really generate a complete screenplay.
Agree about 2019 and stale data, I need the scripts and Box Office and Academy Award winners, all the fresh data. If I can’t get Queen’s Gambit in there, it’s not ready yet! jk
(am collecting modern recent scripts of all kinds, free public domain but also purchasing/licensing proprietary rights)
Depends on which stack you use to build, but these are a good place to start quill, slate, draft-js, or draft-wysiwyg (with various trade-offs). I’m also building my own in my spare time and am open to collabs .
Fair game is right
. Have you tried using finetuned models instead of prompted ones? From my own experiments you can get much much lower latency, as well as higher quality that way. There are also lots of possibilities with optimizing for specific tasks with curated datasets ie tuning the AI-assistant to write spec scripts in the style of A24 horror movies. Also the base models can’t make references to events post 2019 but finetuned models can if they’re trained with the data (ie more fresh and relevant takes).
.