AI takes on Hollywood's tedious tasks
An AI-generated graphical representation of a movie script. Image: Rivet.ai
A new company wants to help people make movies by outsourcing the grunt work — scheduling, budgeting, script analysis — to AI. Starting from a human-written script, its algorithms can draft a budget and a shooting schedule, and even look for plot holes.
Why it matters: Debajyoti Ray, the founder of Rivet.ai, says that AI tools can cut down on uncertainty and allow production companies to take bigger risks rather than re-making the same superhero movie dozens of times.
Rivet.ai spun out of End Cue, the production company behind a trippy, nonsensical AI-written short film called Sunspring. Ray, whose background is in natural language processing, said his philosophy changed after Sunspring: his new company is focused mostly on analyzing human writing rather than trying to emulate it.
How it works:
- In a few minutes, Rivet.ai's software can process a movie script and extract key elements from each scene: the characters involved, the location it takes place in, the props required, and the types of shots they call for. "This is a very laborious process" for humans, Ray said.
- The software can then work out a shooting schedule that takes the availability of locations, props, cast and crew into account, plus weather forecasts.
- Informed by the schedule, the platform can draw up a budget based on compensation, location and prop costs.
Beyond administrative help, Rivet.ai is offering AI-powered advice for writers, too — a more difficult proposition.
- The service can plot the script on a graph that shows continuity between scenes, charting the frequency and tone of various characters' interactions to call out holes in the narrative or extraneous segments. If a scene needs to be removed, the graph can show all the earlier and later ones that will be affected by the altered storyline.
- Most ambitiously, Ray says that the system, trained on scripts available in the public domain, can nudge less experienced writers — corporate PR arms, for example — toward tried-and-true plotlines and structures that are "known to perform better."
But, but, but: Algorithms that pigeonhole stories based on familiar archetypes and estimated performance run the risk of outputting variations of the same script over and over. Other AI-in-film startups like ScriptBook and Pilot help predict scripts' commercial success, automating processes that until recently relied to some extent on human intuition.
Ray argues that rather than sending new films down well-worn lucrative ruts, more data will free up producers to pick up interesting scripts that might not immediately seem commercially promising.
The big picture: Instead of trying to make an AI screenwriter — a task for which technology is not yet suited — Rivet.ai's software aims squarely at administrative work. "Nobody went into filmmaking to do script analysis," says Ray. For now, the creative aspects of filmmaking are still out of reach of automation, though AI-assisted decision-making on things like budgeting can have knock-on effects for an entire project.
Go deeper: The New York Times reports on a startup called Arraiy that uses AI to automate tedious CGI tasks.