Using AI in film casting
Can we develop a machine learning based project to automate the film casting process?
Cinema is undoubtedly the most popular artistic expression in the contemporary world, able to unite millions of people from all over the world under the same passion. Although it is defined as the seventh art, technical aspects coexist in film production, such as management, and purely more artistic aspects, such as the scenic composition or the direction of photography. And while in the first case it is easy to automate work with computer tools, creative flair is not exactly a feature of computers and algorithms. Or at least that was thought until a few years ago.
The increasingly massive advent of artificial intelligence (AI) has made it possible to replace human work in order to speed up and automate even in film production, and not only in administrative areas. In essence, an AI works thanks to a sophisticated algorithm that analyzes and studies the human work and then replicate it to perfection, allowing you to make more or less creative choices and jobs just as if it were a person to perform them. This operation is called more specifically machine learning (ML), precisely “machine learning machines”.
They are currently in the AI research phase that studies human behavior and can replicate it in various stages of film production, such as editing or writing. And although it may seem strange, it is also developing a model of ML able to replace the man even in casting. This project, called Casting Matching Process Artificial Intelligence (CMP-AI), proposes to create the profile of a character analyzing it directly from the script, based on keywords based on three narrative elements: physical-aesthetic, psycho-aptitude and scenic. The profile obtained from the analysis of the keywords is then compared with a database in which the profiles of real actors are inserted, obtained using the same method of automatic analysis, in order to find the actor most compatible with the character in analysis.
More specifically, the profiles of the actors are written using the same three elements used for the profiles of the characters, but in this case instead of the script is analyzed a film (or any other audio-visual format) in which there is the actor to be profiled, inserting in them physical, psychological and contextual characteristics that a particular actor is more or less to interpret and in what way. This makes it clear that the more movies and scenes are examined by the algorithm, the more accurate and realistic the profile of the actor is. All the profiles of the actors then go to compose the database of actors’ profiles, so the more this database is provided, the better the comparison with the profiles of the characters will be.
The results of the comparison are expressed in percentage points, indicating precisely the degree of compatibility between an actor and the character to be played. After the comparison phase, which is in fact the beating heart of the whole system, the program draws up a list of actors from the most compatible to the least, so as to suggest the best actors for the role, but giving the user the opportunity to examine every possible alternative to the first result. There is also a virtual preview mode, rendered in 3D graphics, illustrating an audition for the character played by the actor selected for the demo.
It is thought that this model of machine learning could greatly help film companies in choosing the cast, making the process faster and more practical, therefore less expensive. The most discussed perplexity on the application of this automation, and more generally on the use of artificial intelligence in the cinematographic field, is that sooner or later they will be able to replace human personnel at all stages of production and that inevitably the cinema will become less and less an art and more and more an industrial assembly line process.