Here’s how AI-generated movie trailers spare the editors’ time
OTT platforms, streaming services, and movie producers put a lot of effort into creating stuff to watch — just as much goes into the promotional process to actually get us to watch that stuff.
We can all agree that trailers are where it’s at with promoting a movie or a TV series. They manage to get the most hype and tell the crowd what the movie is about — so everybody wins.
Trailers, important as they are, take up a lot of valuable time for editors. So today we are looking into how we can trim some of that down by automatically generating movie trailers.
We got a problem — manual trailer generation takes too long
Creating movie trailers may seem like a simple job — you just cut some scenes of the movie and put them together, right?
The truth is that a trailer is the flagship of a movie promotion campaign, and that means that every little detail has to be just perfect.
Because of that, editors spend up to a month to create a single trailer, trying to come up with a sequence that teases just enough of the movie without giving away the best parts.
And the time factor is the problem here.
The laborious manual process can take up anywhere from 10 up to 30 days to complete. That’s just one trailer — but usually, studios release several trailers as the movie approaches its release in the theaters.
On top of that, there might be a need for different trailers for different markets. Add that up and you have a lot of work to do.
How do studios approach AI-generated movie trailers?
As of now, the majority of studios continue to rely solely on manual labor when it comes to making trailers. Yet there are concerns about how much time can be saved if we could somehow speed up the whole process.
A couple of years back, tech companies have approached Hollywood proposing to utilize Artificial Intelligence that would automatically select the most salient scenes of the movie or a TV show and cut them into a short trailer.
In 2016 20th Century FOX partnered with IBM to try the technology out. The Watson APIs were tasked with creating a “cognitive movie trailer” for the upcoming horror film “Morgan”.
Would you like to discuss the article?
The first thing they had to do is to teach the algorithm to understand what parts of the movie are important. They have fed the system 100 different horror trailers so it would analyze the following aspects:
- Visuals. The technology learns to identify people, objects, and scenery. The scenes of the trailers were tagged with emotion so that the algorithm could differentiate those as well;
- Audio. Artificial Intelligence analyzes ambient sounds like music or characters voices in order to understand the scenes overall tone;
- Scene composition. The technology looks at the location of the scene, lighting, and framing to learn how horror movies are usually shot.
This way, the AI gets a feel of what scenes make up the horror movie.
Then it was time to put the AI to the test.
The system was fed the entire “Morgan” movie, and the AI has managed to identify 10 tender and suspenseful moments that would make a good trailer.
Eventually, IBM had to bring in the human editors to cut those scenes into a single sequence. Even though the system recognized the salient content for the trailer, it has no ability to cut and compile the footage.
On top of that, the final result still required a polish from an editor. The studios still remain undecided on automatically generated movie trailers, keeping in mind that a trailer as a promotion of creative work requires human-like creativity.
Another issue here is emotion labeling. With this approach, the algorithm has to be trained for each movie genre separately to understand the specific characteristics of other genres.
Apart from IBM, researchers at the University of Edinburgh have tried their hand at AI-generated movie trailers.
They have built a neural network that can generate movie trailers based on the unsupervised, graph-based machine learning algorithm.
They have started with breaking down the task into two separate sub-tasks: the analysis of the movie’s narrative structure and the mood it conveys. The algorithm is given the whole movie to analyze, as well as the extracts from the movie’s screenplay.
The researchers had two separate neural networks going through each sub-task and combined, they were able to find and identify the major turning points of the movie, such as a change of plan, a setback, the point of no return, or a climax.
The AI was further tested on 41 more movies. The resulting trailers were compared to the footage produced by supervised learning models, and polled viewers seemed to prefer the first group of trailers.
But nonetheless, some of those solutions go as far as finding the trailers-worthy content within the movie and do not generate the footage automatically. Even with such technology implemented, the studios would still have to hire editors to polish the scenes and compile them into a trailer.
There is a better way to optimize content production.
Another problem here is that some algorithms require additional data sources apart from the movie to pinpoint the best moments for an AI-generated movie trailer, such as an extract from a screenplay. That may work well enough for movie studios, but what if an OTT platform or a streaming service wants to get some trailers for their lesser titles? Those sometimes may not have a trailer of their own, and platforms are not provided with the script outline to feed the system.
So what can we do about this?
How we solve the problem of AI-generated movie trailers
We rely on advanced cognitive computing technology to deliver top-notch video-processing solutions for media and entertainment companies of any scale.
Our products possess a human-like understanding of context to provide the required results without the additional datasets in a matter of minutes.
When it comes to creating movie trailers, CognitiveMill™ analyzes video content and selects the highlights of the movie as the human editor would. The accuracy is the same: the great thing is that you get your trailer in minutes.
The tool decomposes the visuals to understand the context of the movie, analyze its plot and identify the most trailer-worthy moments, according to the movie’s genre.
Get the best AI-generated movie trailers out there
Artificial Intelligence found its use across many industries, with media and entertainment companies also looking into it as a way to optimize their workflow and automate some of the mundane tasks.
Now, the production studios, OTT platforms, streaming services, and broadcasters can enhance their movie and TV show promotion by automatically generating trailers: whether you want to create a trailer for those movies without the official one, or generate a couple of different trailers to cater to several markets — it all can be done much faster.
CognitiveMill™ tops the competition when it comes to details of AI-trailer generation:
- CognitiveMill™ doesn’t need additional data sources. The solution pulls all the data from the footage itself to get a feel of the context. It does not require extracts from a movie screenplay or something of that sort;
- The solution provides superb accuracy. Cognitive computing complex algorithms allow the tool to thoroughly analyze the plot of a particular movie to get the gist of what’s going on. To complement the scene breakdown, it applies audio analysis to deliver the most precise selection of trailer-worthy moments;
- The tool completely automates the process. CognitiveMill™ not only finds the best moments of the movie but also cuts and compiles them into a finished trailer. Since the tool is tied into the ecosystem of other products, you can also automatically share the trailers on social media, or adapt them for mobile viewing by cropping the footage from landscape to portrait aspect ratio.
Though we have managed to reach a high level of accuracy, it is still pretty hard to match the creative potential of a human editor cutting the trailer.
We are currently working on improving our first iteration of the tool to introduce a more character-driven approach to scene moment selection. On top of that, we will be adding the ability to set parameters for trailer generation to offer a more agile experience.
Interested in setting up an automated trailer generation workflow? Reach us at support@cognitivemill.com or fill in the contact form below. We will provide more information on the product, schedule a demo, and discuss the cooperation terms!
