Project Greenlight

Greenlighting a movie is a risky proposition. What if there were a tool that read through unproduced scripts, found how unique it was and what its box office return may be?

Comment

Inspiration

With the success of films like Obsession, the movie world has been put on notice that unique scripts, and not necessarily blockbusters made from existing IP or sequels, can be insanely profitable. As always, the story matters, and finding that unique script can be a highly profitable exercise.

What it does

Project Greenlight takes an original screenplay in PDF format and scores how unique and viable it is by comparing it to a database of over 18,000 movies. Similar movies (based on plot and/or similar genre) are then weighed and used as a basis to determine the screenplay's score across these metrics (from 0 to 100, the higher, the better):

  • Commercial potential: how much money can it make?
  • Originality: how unique is this plot?
  • Audience/genre demand: are people clamoring for it? This one is based on how user scores of similar movies and movies in the screenplay's genre(s).
  • Production feasibility: how hard is this to make?
  • Evidence confidence: how certain is the model of these metrics?

It then presents a sample of similar movies and the screenplay's potential gross (earnings vs budget).

How we built it

This app was build on Replit with a Google Gemini 2.5 agent taking care of parsing the screenplays and determining its metrics. I used a QDrant vector database for the semantic searches and a MySQL database (hosted on Google Cloud) to feed the vector database and then to obtain budget and box office total information. I also used a dataset from Kraggle containing over 22,000 movies (with information from tmdb and imdb), which I trimmed down to a little over 18,000 by removing those with missing plot, genre or other important information.

Challenges we ran into

Creating the QDrant database was a massive undertaking, since the MySQL database kept timing out. Permissions also turned out to be an issue, since the IPs kept changing on the Replit side and I could not get the connection to MySQL to hold steady.

Accomplishments that we're proud of

The ability to parse scripts and not only get a summary, but also to be able to compare it against previous movies and figure out just how unique (or derivative) the idea is, was a really cool exercise, and I ran quite a few unproduced screenplays to see just how viable they are.

What we learned

AI can be a powerful tool when it is fed hard data. The fact that Kraggle had a dataset with that many movies really helped the LLM determine the uniqueness and projected gross of each potential film. I also learned that Replit is an amazing tool that drastically cuts down development time and is a great partner to bounce ideas off. I will definitely pair up Gemini with Replit in future projects.

What's next for Project Greenlight

It would be awesome to actually have the application be part of a movie studio's daily jobs. It would scrape potential scripts, run the pipeline and if the script has potential (let's say a high uniqueness score and/or high box office potential), then it automatically contacts the studio's decision makers so they can take a look at the scripts. This would eliminate the "noise" of all the derivative scripts and let the unique and potential high-grossing scripts actually get in front of the people with power.

Built With

Share this project:

Updates