Researchers at 20th Century Fox created an AI tool called Merlin Video that can predict movie audiences based on trailers.
The research team trained a deep learning system using data that linked hundreds of trailers to movie attendance records. Thanks to the training, the AI can connect visual elements with move performance among demographics.
Merlin Video can look at trailer elements like colour, faces, landscapes and lighting.
Furthermore, the AI tool’s predictions are accurate. The researchers tested the AI against existing movies, where it performed well. However, the researchers also had success predicting future movie performance.
Additionally, the team said this is the first instance of using AI to predict customer interests from a movie trailer.
The AI method can make predictions based on text or video. Text predictions compare plot points, whereas video predictions are based on visual elements. Ideally the researchers would like to combine video and text to get a fuller sense of the story.
Regardless, the company is excited about the practical applications. Trailers form a vital part of movie marketing. The ability to tailor a trailer to appeal to an audience would be beneficial.
However, such methods of predicting interest could lead to samey trailers. If highlighting specific visuals leads to greater interest from customers, it could mean seeing similar visuals in a number of trailers.
Ultimately, it will be interesting to see if and how this tool affects future trailers and the movie industry.
Fox’s seniour vice president of data science and analytics, Miguel Campo-Rembado, authored the paper with his team. It was published on the ArXiv server ad can be read here.