Some two years ago we published an article that tried to answer the question of when will we see the world’s first AI movie director?
In this article I stated that “It seems certain that the first AI directed movie will be either a cartoon or animated film.” Well, two years on, it was brought to my attention, rather embarrassingly, that some people feel that by the time I even wrote that article that we had already seen the first AI directed film.
I noted in that article that the “first ever AI generated script that was turned into a movie called Sunspring (2016)”, a real milestone for artificial intelligence. Well, it turns out that it was an evolution of the same artificial intelligence system that, only two years later in 2018, did in fact ‘direct’ a movie.
The film that this AI system, which its creators called “Benjamin”, wrote and directed was called “Zone Out.” Benjamin used its previous training to, where it was fed 100+ film scripts in order to teach it to write its own script and to create this new script based on the following perimeters:
Title: Zone Out
Dialogue: ‘They wanted to call it Adam – seriously”
Prop and Action: A character holds a lens and turns it to reflect a bright light
Optional Science Idea: A genetically tailored virus targets only pregnant women
But now Benjamin faced two new challenges – acting and directing.
Fortunately, once its original training to create movie scripts, Benjamin had been trained to both act and ‘direct’ the film too.
Starting with the acting, Benjamin was trained to interpret video footage of old films in order to ‘understand’ acting in a more comprehensive way. Interestingly, the AI engineers training Benjamin also included “Sunspring” in order to analyze the performances of actors Thomas Middleditch, Elisabeth Gray and Humphrey Ker as they performed the script that it had written. Other films included classics such as Night of the Living Dead, The Brain That Wouldn’t Die and The Last Man on Earth and so on.
Once it had a better understanding of what the scenes and various performances represented, the system then used facial swapping technology and voice generators to create new performances by modifying prerecorded clips of the actor’s faces and voices.
The system then learned how to “direct” a new film by slicing clips from the films it had analyzed together and combining it with the acting modifications mentioned above.
Finally, the system assigned a musical score to the soundtrack and 48 hours later the film was done.
You can watch Zone Out via this link to see the results.
While AI and ML technology has an enormous way to go, the results do give some idea of the promise of artificial intelligence to help in the movie production process.
The limitations of AI really mean that when a creative AI director does emerge, one who is able to think ahead of the create curve in human terms, then it will most likely require more human input than their human director counterpart does.
The reason for this is that all AI systems require teams of AI engineers and data scientists in order to train and maintain them. All of the current AI-assisted moviemaking platforms, for example, have teams of highly trained professionals behind the scenes who work tirelessly to develop the system on a day by day basis.
Early systems will require a full-time director on hand to interpret the insights of the system.
Since most of us don’t have much understanding of how AI systems work and how to use them, companies need to have professionals either on site or online to deal with customers queries. In the physical world where most of current movie production is undertaken, this means engineers on site to help train and interpret the outputs provided by their artificial intelligence systems and to work alongside the existing director.
Indeed, should AI ever truly be accepted into the role of directing a film, and indeed, be able to produce work that would be accepted by audiences, the physical limitations of it being digitally based and not being able to physically control a camera, etc mean that the movie business will need human input for a very long time to come.
Again, I fall back on my prediction made in my original article that AI will most likely make a huge splash in the animated film genre, where eventually, individual versions of movies could be generated by an all in one artificial intelligence system for individual audience members or specific cultural demographics (think language and cultural elements such as politeness, etc) and released in a matter of hours. Such ‘customization’ would make films more popular and, as a result, successful.
For more ‘traditional’ films filmed in the real world with real actors, it seems unlikely that AI technology will be taking control of movie production any time soon. While animation is a popular genre, especially with children, it is yet to be seen if we humans will accept computer generated images featuring computer generated stars over our current human ‘stars’.
In the short term, it is likely that the most helpful use of this technology will be in the video game industry. AI directors would be able to tailor the game to the individual player to build in uniqueness and, therefore, a greater appeal to its games.
An early example of this is AI Director 2.0 that is used in the ‘Left 4 Dead’ multiplayer game that varies the game by deciding where a player will morph into the game, etc.
This technology is easily applicable to computer games that have no ‘physical’ elements such as human actors etc. Indeed, it is likely that this is where the short term money will come from for this development of AI director development.
As we all know, money directly influences the speed of development when it comes to technology, so it is very likely that until ‘AI directing’ technology becomes a cornerstone of interactive computer games, it is very unlikely to be replacing the human movie director any time soon.