Tristan GreeneEditor, Neural by TNW
Tristan is a futurist covering human-centric artificial intelligence advances, quantum computing, STEM, physics, and space stuff. Pronouns: Tristan is a futurist covering human-centric artificial intelligence advances, quantum computing, STEM, physics, and space stuff. Pronouns: He/him
You can have Citizen Kane and The Godfather. Keep your 3D films and IMAX experiences. Miss me with epics, sequels, and experimental cinema. The only real film-based art form is movie trailers.
When you go to a theater there’s no guarantee you’re not about to sit through a 90-minute stinker. That’s the chance we all take when we fork over $25 bucks to see this year’s blockbuster sequel.
But, for my money, it’s always worth it as long as I get there early enough to enjoy the previews. Which is why I’m exceedingly excited over this pre-print research paper from the University of Edinburgh.
According to the researchers, they’ve developed an AI model based on a pair of neural networks that’s capable of spitting out a fitting, emotionally-captivating trailer for any given movie.
In fact, the team used the system to spit out more than 40 trailers for existing movies and Amazon Turk workers tended to prefer the AI-generated trailers to the official ones.
Per the team’s paper:
To create trailers automatically, we need to perform low level tasks such as person identification, action recognition, and sentiment prediction, but also more high-level ones such as understanding connections between events and their causality, as well as drawing inferences about the characters and their actions.
They were able to accomplish this by utilizing two separate neural networks. The first processes the film’s video and audio in order to, essentially, find scenes of interest.
And the second is basically the arbiter of what’s interesting. It views a textualized version of the film – similar to a screenplay – and uses natural language processing to find important and emotional moments.
The completed model generates novel trailers using “movie understanding” based on how the neural networks process the input data.
Previous work in the field includes a collaboration between researchers working with IBM who built their own bespoke movie trailer-generator. But that one apparently only generated trailers for a specific horror film.
Typically, AI generators are trained on thousands of hours of similar content in order to show the machine what it’s aiming for. But these models aren’t trying to hallucinate something that looks like a movie trailer, they’re actually stitching together one. That’s something that involves more than just tweaking parameters inside a black box.
A good trailer can make or break a movie, financially speaking. And an AI that can democratize the creation of high-quality movie trailers could be a game-changer for independent filmmakers.
Not to mention the fact that I’d rather watch 90-minutes of AI-generated trailers than most movies anyway. Here’s hoping this finds a web release some day so we can try it out for ourselves.
Get the TNW newsletter
Get the most important tech news in your inbox each week.