Stuffmesh.com is reader-supported and the following article contain affiliate links, When you buy through links on our site, we may earn an affiliate commission.
Editing movie trailers is an art. Apparently, those who shoot, direct and edit these films often do not produce the trailers, which are supposed to put the butts of all of us in cinemas (or at least some of us these days) – and those who do are outside companies that specialize in the art of the “cruel world” . “One man is fighting against all odds.” But, as in quite a few other fields, the field of trailers is beginning to creep into the field of trailers and is armed with determination and neural networks.
The model who studies the film, recognizes what is important and produces a trailer
Researchers at the University of Edinburgh in Scotland have developed an artificial intelligence model built from two neural networks, which can take all the visual data of a particular film and produce a trailer that will not fall short – and may even surpass those made by a professional editor.
In an article published by the researchers, the model they created was able to produce the trailers after performing some basic actions – such as identifying the characters, identifying the actions each character does and being able to try and predict the motivation behind each character’s action in the film he examined.
In addition, the model – after running and examining the film – understands the connections between the various events that happened in the film and their consequences, and draws conclusions about the actions taken. The model, as mentioned, is made up of two neural networks: the first network is the one responsible for the video analysis, in order to find scenes that can be useful for the benefit of creating the trailer. In fact, it works similar to the impressive Video Analyzer tool that Microsoft is developing here in Israel.
The second network is the one that determines what will interest the viewers and may attract them to watch and what not. To do this, the researchers feed the model a textual version of the film – something similar to the script, but presents the plot of the film more comprehensively on the textual level. By using natural language processing, the model recognizes the most important moments and scenes in which a lot of emotion is expressed, a necessary component of a good trailer, and marks them.
The researchers write that the model they developed produces the trailers using what they call “understanding the movie” of neural networks, and based on this understanding the model processes the information – which is actually the movie – into a trailer that should be more interesting and successful than humans.
The model developed by the researchers was mastered by watching more than 80 different movies, which contain more than 11,000 different scenes – and for them the trailer studios created about 280 different trailers. The model was then tested by using 38 films and created a total of 40 trailers for them.
In 2016 there was a similar use of the technology by researchers from IBM, who teamed up with 20th Century Fox movie studios and also created artificial intelligence in favor of creating trailers. However, the same model that was developed produced a trailer for only one film – a horror thriller called “Morgan” which was then released in cinemas.
In the results of their research, the researchers note that the 40 trailers created by the model they developed were also humanized by Amazon Mechanical Turk, the US giant’s outsourcing platform, who tested the trailers, where there was a clear preference for trailers created by AI over professional studios.
In any case, the researchers point out a very important point. Their model, smart and useful as it may be in creating interesting and engaging trailers, does not yet know what spoilers are. So it’s too early to think about him editing the trailer for Marvel’s next film, because he could easily reveal some dramatic points that might have been nice if they had surprised us all on the big screen.