IBM Watson Created The First-Ever AI-Made Movie Trailer For 'Morgan' (popsci.com) 58
An anonymous reader shares a Popular Science article: For a film about the risks of pushing the limits of technology too far, it only makes sense to advertise for it using artificial intelligence. Morgan, staring Kate Mara and Paul Giamatti, is a sci-fi thriller about scientists who've created a synthetic humanoid whose potential has grown dangerously beyond their control. Fitting, then, that they'd employ the help of America's AI sweetheart IBM Watson to build the film's trailer. IBM used machine learning and experimental Watson APIs, parsing out the trailers of 100 horror movies. It did visual, audio, and composition analysis of individual scenes, finding what makes each moment eerie, how the score and actors' tone of voice changed the mood--framing and lighting came together to make a complete trailer. Watson was then fed the full film, and it chose scenes for the trailer. A human -- in this case, the "resident IBM filmmaker" -- still needed to step in to edit for creativity. Even so, a process that would normally take weeks was reduced to hours.
Good trailers? (Score:1)
Re:Good trailers? (Score:4, Insightful)
Everybody thinks everybody else's job is easy.
Re: (Score:2)
Re: Good trailers? (Score:1)
But those things wouldn't be expedited by an A.I. making the trailer...
Re: (Score:3)
Re: (Score:1)
I certainly know people who do. You might as well ask why it takes so long to write a book. Or code. It's just text, you know?
Re: (Score:2)
Slapping together a trailer of scenes from a movie isn't that different from slapping together a quick little program combining libraries someone else already wrote. It's not like the trailer maker has to go out and do a new shoot, edit for lighting and audio, add CGI effects etc. He LITERALLY has to watch the movie and at a few points go "Ooh, that looks awesome, that goes in the trailer!"
Re:Good trailers? (Score:5, Insightful)
Hmm, no.
1 - you don't want your 'best' scenes in the trailer or people will go 'that film was shit, just watch the trailer'
2 - you don't want plot spoilers in the trailer
3 - your trailer will be targeted at specific audiences. Different elements from different scenes will appeal to different audiences
4 - you want to convey specific messages with a trailer, and so its creation is an inherently creative task to assure this
5 - trailers can be cut and edited differently to the film. E.g. cutting the middle from a scene to hide a plot spoiler
6 - trailers can use scenes not included in the film. Some directors shoot scenes for the trailer
7 - trailers need to make narrative sense in their own right. Shit, even people posting crap on Youtube do more than post a sequence of scenes from a film
8 - trailers can fuck with a film's running order to support the above
That's before you add in a trailer voiceover, interstitial text/images, the need for trailers to provide consistency between Cinema (three minutes of trailer) and TV (30 seconds), the need for trailers to keep consistency with other marketing and PR (cast interviews, written reviews, press releases, 'fluff' pieces, astroturfing, all the other shit the media industry do).
But sure, just fucking slap together a trailer. Go for it. You clearly have a sparkling career ahead of you.
Re: (Score:2)
Slapping together a trailer of scenes from a movie isn't that different from slapping together a quick little program combining libraries someone else already wrote.
npm install scene --save-trailer
Re: (Score:3)
Everybody thinks everybody else's job is easy.
I take it you've never seen Dirty Jobs with Mike Rowe.
Re:Good trailers? (Score:5, Funny)
Not really. I think the guys around the corner building a highly efficient and flexible network stack have a tough job.
Making a trailer isn't that difficult. Just take all of the scenes that detail every bit of the plot and slap them together. Once you have something that removes any potential chance of enjoying the film then it can be called complete. In some cases, when the film is so incredibly terrible, it may be necessary to fabricate a trailer that has barely anything in common with the film. Sure, you probably should have made that movie instead, but then we wouldn't have done all of this blow.
Re: (Score:2)
Ah, I see you graduated from the same school as both the Ghostbusters and Suicide Squad trailer creation teams.
Human Editor Helped (Score:5, Insightful)
The trailer was alright. Given a human helped edit what Watson came up with, I wonder how much work they did or how different the final trailer is from Watson's trailer. Given how formulaic film trailers are already, it makes sense that Watson could learn this pattern and then reproduce it given footage from the film; ensuring that the scenes used hint at a premise without being straight-up spoilers would probably require a human editor.
That said, the film currently has a 48% rating on Metacritic, and apparently isn't that great from a sci-fi or thriller point of view.
Re: (Score:1)
"What Watson did was analyze the stream and find the "action" scenes."
i.e. it looked for small time intervals between key frames and chose a selection of 100 of those.
I think IBM are keeping Watson away from actual AI stuff, lest we know how crap they are. This way they can pretend to be leaders in AI, patent a lot of fluff and make money leaching off the real AI companies doing self driving cars, image search, and so on.
Here they can pretend to be capable of video comprehension, which will come in handy wh
Re:Human Editor Helped (Score:5, Insightful)
I think IBM are keeping Watson away from actual AI stuff
Who cares? The real danger here is that they let the most advanced AI machine on Earth watch tons of horror movies. Am I he only one who finds that a bad idea?
Re: (Score:2)
Doesn't matter because IBM Watson isn't remotely intelligent, it has no idea what it is doing, it doesn't make choices, it's just an advanced database with 'neural net's thrown in. Data in, data out, nothing more.
Pattern Recognition (Score:5, Insightful)
Its a little disingenuous to say that Watson "created" the trailer. The only thing Watson did was run a pattern recognition algorithm to figure out which clips in the movie were tense, happy, scary, etc. Then a human editor sorted through all of the clips, picked the good ones and put them in sequence to create a trailer that actually had narrative instead of just being a hodge podge of disjoint clips.
Pattern recognition is getting better which is the first step to creating an AI... but Watson, and AI in general is still very far off from creating a computer program that is capable of original thought.
Re: (Score:3)
It would be very interesting to know how much human editing was done on the final trailer.
Well, from the Youtube video description:
Utilizing experimental Watson APIs and machine learning techniques, the IBM Research system analyzed hundreds of horror/thriller movie trailers. After learning what keeps audiences on the edge of their seats, the AI system suggested the top 10 best candidate moments for a trailer from the movie Morgan, which an IBM filmmaker then edited and arranged together.
Obviously this is just a brief explanation, but to me this implies the "Ai system" did little more than suggest 10 moments from the movie to string together. The human editor then actually did all the work of editing the clips, choosing the order, etc. That implies a HUGE amount of freedom on the part of the human -- if I can use a 1-second clip or a 20-second clip from a particular "moment" (perhaps edited as many movie clips in trailers are?) and I can put the or
Re: (Score:2)
"Recognising tense happy and scary is interesting in itself."
It didn't do that. The computer analyzed 100s of horror movie trailers and then selected parts of the movie that where the similar to the trailers.
"It would be very interesting to know how much human editing was done on the final trailer"
After the computer selected bits of the movie, which of those bits were used, edit points, and everything else was done by humans.
Re: (Score:1)
Re:Pattern Recognition (Score:5, Interesting)
The studios subcontract trailers out to the trailer houses, usually multiple ones for each film. Then they micromanage the crap out of the process at each house. Then they sometimes take all the trailers and re cut in house to come with a Frankentrailer that some committee agrees on. That's often the best case scenario.
My son has been made to make up dialog that's not in the movie at all before. The talent couldn't be gotten to record something so he was instructed to create a line from phonemes cut and pasted into a file. I kid you not. Sometimes a studio will bring in the dailies and have him start picking stuff knowing that the actual film might not even have the scenes after the final edit.
And then there is the fact that many trailers of different length and content are made for different markets and media.
Re: (Score:2)
That explains this story [slashdot.org].
Bringing things to the extreme... (Score:5, Funny)
So humans will have more time to do something interesting, live viewing old b/w movies at home, while robots are elsewhere at the movie...
yeah... (Score:1)
bullshit.
So what's the big deal? (Score:2)
I watched this trailer for "Morgan", and I gotta admit - I just don't get it.
It was just a bunch of people standing with one knee in the air.
Re: (Score:2)
I watched this trailer for "Morgan", and I gotta admit - I just don't get it.
It was just a bunch of people standing with one knee in the air.
lol, I think I'm the only one who got that joke.
Misleading (Score:1)
It is misleading to say 'IBM Watson' did it. Watson is nothing but a bunch of cobbled API that do nothing that could not be done with APIs from other vendors.
Notably, Watson is not 'intelligent' and the APIs do not include any off-the-shelf conversational agent you can talk to, or for that matter any creative component
that could produce a movie or trailer. Sure, it's good marketing, but it's dangerous to succumb to that imprecise hype, because it creates unreasonable expectations
about what's possible.
Kudos
Anyone still wondering... (Score:2)
... why all the recent movies are basically the same thing with slightly different skins?
Here's your answer.
Captured it (Score:2)
Harnessed reverberation, bass voice and predictable plot the US audience loves so much.
AI for brilliant mediocrity. Yay!
Resistance is Futile (Score:1)
Humans? Who needs 'em? (Score:4, Insightful)
Greatly disappointed by the lack of humorous or insightful comments, but maybe they exist without visibility or sufficient positive moderation. Not surprised, but that's how slashdot has evolved. I don't see the joke in the topic, so I'll go for insight, such as it was. Kudos to you if you can see a joke in this topic.
This article is a good example of how Watson can be leveraged to cut out the humans. Where the studios used to need a significant number of people to do the work, now a single editor with this leverage will probably be able to keep up with all the movies they want to make, and use the leftover time to make more trailers to pick from for each movie, to boot.
I think it's part of the big secret plan of today's so-called IBM. "Respect for the individual" was the OLD idea. The new goal is "Replace the individuals". A few transient employees brought onboard as briefly required. The only question will be "Who can do the work most cheaply?" Actually, there might be one question before that: "Can Watson to do this work without needing any human being?"
Re: (Score:2)
Greatly disappointed by the lack of humorous or insightful comments, but maybe they exist without visibility or sufficient positive moderation.
You may not realize this, but 90% of Slashdot stories and 98% of commenters have now been replaced by AI. Humor & insight will come along in future versions.
Re: (Score:2)
I think that deserves a funny mod (on the theory you are in the other 2%), but I haven't had a mod point to give in many years.
Re: (Score:1)
Humor & insight will come along in future versions.
We're half way there [youtube.com]
Solsbury Hill (Score:2)
Repeatable? (Score:2)
Only the one question then: what happens when Watson is fed the film a second time? Does it produce the same trailer as the first time?
If so, it's totally useless.
No one's interested in an algorithm to creativity. It quickly becomes an expression of nothing at all. I believe we used the term "formulaic" to insult any such "creative" process.
And no, random different isn't any better.
This is intended to be art, people. If it doesn't express creativity, then it simply isn't art.
I'll give you that Watson, a