Movie Studio Lionsgate is Struggling To Make AI-Generated Films With Runway

An anonymous reader shares a report: Last year, the AI video company Runway joined forces with the major Hollywood studio Lionsgate in a partnership the pair hoped would result in AI-generated scenes and even potentially full-length movies. But the project has hit a snag. According to a report by The Wrap, the past 12 months have been unproductive. Lionsgate distributes Hollywood blockbusters including The Hunger Games, John Wick, The Twilight Saga, and Saw franchises. But despite its huge catalog, it is simply not enough for the AI to produce quality content.

"The Lionsgate catalog is too small to create a model," a source tells The Wrap. "In fact, the Disney catalog is too small to create a model." Despite Runway being one of the leading names in AI video, the technology needs a copious amount of data to produce AI-generated films. It is the reason AI has proven to be such an unpopular technology, as AI firms help themselves to any type of media they can get their hands on -- whether it has copyright protections or not. Another issue is the rights of actors and the model for remuneration if their likeness appears in an AI-generated clip. It is a legal gray area with no clear path.

The Human Brain...

[zurkeyon]

Can store approximately 3 Petabytes. (3 Million Hours of SD Video) And as such, one must consider that you need about this much capacity and sampling for accurate pattern recognition and to formulate a proper and developed "Imagination"

Re:

[Fly Swatter]

You give current AI way too much credit for even being close to anything like an imagination. It can't even regurgitate facts most of the time.

Re:

[zurkeyon]

That IS what I said. AI has nowhere NEAR the capacity and sampling ability that the human brain has. And even larger sample sets still "Hallucinate" often ;-)

Re:

[allo]

Arguably, the SNR of the brain is worse than that of a model trained only on the information one is interested in. That's why most image AI can create images that most humans cannot create. It was trained on images and images only, while a human brain is a generalist.

Re:

[zurkeyon]

But when the data amount needed to train said model is unavailable, and unlikely to exist for another 50-100 years, they are incapable of "Imagining" what it is they are being asked to draw, and tend to do nothing but hallucinate gibberish. Their capacity for true decision making remains near zero.

Re:

[allo]

I'm not sure where you see the problem. There are existing image models that create better images than most humans can draw. They are trained on way less data measured in bytes, but also data with way better SNR than what humans are trained on. If you could train your brain solely on images, you could also draw in hundred different styles as fast as your hands can move, but alone the signal from your eyes is way noisier than an image field (but on the other hand you get a constant high bandwidth stream of t

Re:

[zurkeyon]

The six fingered mutants AI regularly produces, beg to differ. AI is nowhere close to being able to filter, process, and rationalize the sheer amount of data human's do. Without the majority of the data that comes from all 5 senses, they are merely pattern recognition drives, with virtually no ability to make "Sense" out of the data beyond the fixed programming of the algorithm that drives their "Choice" Its a clever illusion. But task them with anything real, and its immediately exposed. For now.

Re:

[allo]

Sorry, but I think you should really try a modern image AI to get rid of the three year old stereotypes.

Re:

[zurkeyon]

LoL. Stereotypes? Am I being AI discriminatory? ;-D There is a thousand sites were I can go generate a 6 fingered mutant, or a Crowd with 50 mutant faces, every minute of the day for all time...

Re:

[allo]

The stereotype is that AI cannot do hands. And that's quite outdated. How old is the last image generator you tried?

Re:

[yababom]

Movies are judged by humans against their 'noisy' dataset. If you were able to purge all the non-movie 'noise' from a movie actor/director/writer's experience, they wouldn't be able to make a movie that real humans believe and enjoy.

Re:

[allo]

I agree. And there we're at the point where I see the real limitation of AI: Creativity.
Currently AI can do the craft really well, but you need to steer it. Without putting your creativity and imagination into it, it gets repetitive, for exactly that reason. Given that a brain is also deterministic (on a really low level like chemistry or even atomic interactions), one can wonder where human creativity comes from, but my assumption is, that all the noise in your every day data accumulates to new impulses th

Re:

[sudonim2]

No, it doesn't. Your brain isn't a computer. It doesn't "process" anything. It holds not data. That's not how it works.

Further, we don't actually know how the brain stores memory in the first place. But we do know that human memory fundamentally is a creative process; most of what you "remember" is made up by your brain on the spot as it is remembering something.

Good

[IWantMoreSpamPlease]

Hope it crashes and burns in a spectacular fashion, and serves as a warning sign to the rest of Hollywood.

Eventually AI will win

[Tablizer]

We are at the Model T stage of generative AI. It will get much better, probably either better than CGI, and/or will help create the CGI models themselves which are easier for humans to fine-tune than direct AI output.

However, there will be growing pains. This project is a guinea pig, and is squealing bigly right now. There will be AI roadkill. Gloating about failures may make one feel better, but won't stop the inevitable march of AI progress. The cat is out of the bag.

Re:Eventually AI will win

[evanh]

No, we're at the Philosopherâ(TM)s Stone stage. It's all fumbling wishful guess work. There is no science done yet. It's all a big waste of money.

Re:

[Tablizer]

But current AI mostly works, whereas an alleged philosopher's stone did absolutely diddly squat when observed.

Do you actually believe AI is stuck in its D+ state for say 50 years?

Re: Eventually AI will win

[firewrought]

It's definitely that we wrestle with the pending consequences of AI (both personally and societally) now rather than burying our heads in the sand. However, nothing about the march of AI has a set itinerary... historically the field has cycled thru stagnation and sudden jumps forward. ChatGPT and Dall-E got people's attention and spurred a bunch of investment, but there's no way to know if the next jump forward is happening tomorrow or if we're in for twenty years of mild incremental improvements.

Re:

[SoftwareArtist]

The Model T provided affordable transportation. It worked. It did what it was supposed to do, and did it inexpensively and fairly reliably.

That is not the stage generative AI is at.

Re:

[Tablizer]

Early Model T's broke your arm when the starter backfired. People often bought after-market auto-starters, and then Ford eventually included it as part of the car.

Re:

[phantomfive]

Practically speaking, it doesn't need to be good. People watch cartoons, even though the quality of some of them is outright horrible. No one watched the Simpsons because of the realistic animation.

Re:

[Tablizer]

Good point! Maybe the first successful AI movies will be comedies and spoofs: "Pirates of the Mar-a-Lago"

Re:

[jonadab]

Eh, I kind of hope Hollywood goes all-in on AI generated content, tbh. They haven't produced much that's worth watching any time recently anyway, and if they go under, maybe it'll clear the way for better content creators to rise to prominence, maybe even someone who can figure out how to write a script from scratch, that is NOT the eighty-third sequel to a mediocre nineties action movie, or the twenty-seventh reboot of a superhero franchise.

Re:

[jonadab]

Eh, you're preaching to the choir. I haven't watched a Hollywood movie made in the last fifteen years. The last one I saw in a theatre was LOTR:ROTK. I've seen a couple of movies more recently than that, but they were old ones.

When the thirty-second trailer looks boring and heavily derivative, it's pretty difficult to imagine that the actual movie could hold my interest for over an hour. Sorry, not interested. Do you know how long it's been since I saw a movie trailer or advertisement, that made me wan

Re:

[jonadab]

Although, hmm. I might potentially be persuaded to watch Encanto at some point, if the right other people are watching it and invite me to watch it with them. If so, it'll be the first Disney movie I've seen since Shrek.

Re:

[thrasher thetic]

Seconded. May their efforts be rewarded with frustration and failure.

Re:

[DarkOx]

Why because you like paying $15 for matinee ticket?

You like watching the Nth remake of whatever because productions costs are so high big studios are generally afraid to try things that are different?

The wonderful thing about all forms of 'fiction' where AI is concerned is that it does not have to be "correct" it is after all already fictitious. If AI can write a more entertaining book, or make a better movie, how come we should not want that?

Google indicates SAG has like 170K members world-wide. Obviously

Coming soon

[VampireByte]

Twilight Saw Games

Re:

[sconeu]

Which games did Twilight see?

Re:

[VampireByte]

Ask AI

Re:

[EvilSS]

Twilight characters meeting creative, gore filled ends in a game format? I'd pay to see that.

Maybe the data is not the problem.

[Fly Swatter]

Their algorithms are just not capable.

No one asked for an AI generated movie anyway.

"It is a legal gray area with no clear path"

[hyades1]

From TFA: "It is a legal gray area with no clear path."

Translation: "When you do it, it's piracy. When we do it, it's research."

I am shocked, shocked to find gambling here

[gurps_npc]

AI is not smarter than a human. It is much, much dumber.

Fools think it is smarter because we teach it to specialize in one specific task that it easy for software to do.

It is like thinking a dog is somehow smarter than a human because it can smell drugs.

Our current AI works (on the tasks it is expensively trained to do well) about as well as an intern (on tasks the intern is simply told to do).

It lies, ignores simple instructions that were not part of the training, and generally fails except on very specific tasks.

Re:

[Gilmoure]

It's great at data surveys and filtering huge swaths of data.

They don't need AI

[newcastlejon]

Lionsgate distributes Hollywood blockbusters including The Hunger Games, John Wick, The Twilight Saga, and Saw franchises. But despite its huge catalog, it is simply not enough for the AI to produce quality content.

Seems to me that Lionsgate don't need AI to churn out soulless, repetitive slop; they've been doing it for years.

Re: They don't need AI

[anonymouscoward52236]

They'd rather churn them for $1 each.

Exit Strategy

[xanthos]

"The Lionsgate catalog is too small to create a model"

Translation: We assumed this would be easy, it's not, and this is the excuse we came up with to justify all the investor money we have burned.

Re:

[Gilmoure]

Liongate VP: "Why does every guy look like Keanu Reeves and every woman Jennifer Lawrence?"

GIGO

[Hentes]

Lionsgate distributes Hollywood blockbusters including The Hunger Games, John Wick, The Twilight Saga, and Saw franchises. But despite its huge catalog, it is simply not enough for the AI to produce quality content.

Garbage in, garbage out.

One day

[ArchieBunker]

These studios are going to discover public domain works like in the plot of Reckless Kelly. https://www.imdb.com/title/tt0... [imdb.com]

Re:One day

[White Yeti]

Legal training material will usher in the golden age of AI silent, black-and-white film!

Re:

[ambrandt12]

My guess is, that to avoid the copyright questions, they'll feed the movie-LLM with everything public domain/not copyrighted related to movies, same thing with music, even take an image of Leo DiCaprio and change his face enough that it's not Leo anymore... anything that they don't have to pay for or can wriggle around existing laws is fair game... so, might want to lock-down your Google Drive/OneDrive if you have links out there anywhere to files.

Re:

[ambrandt12]

Well, the story of Ned Kelly is public domain, but the script of Reckless Kelly isn't.

AI Training is Intellectual IP Theft

[atrimtab]

That still does not work reliably.

Is it fixable?

So far it appears not to be...

Of course, part of the issue is that the definition of "what is AI" is purposely vague and flexible marketing....

AI video needs a copious amount of data

[Mirnotoriety]

There you have it. Despite claims to the contrary by the AI companies AIs are useless without other peoples works.

Cataloging excuses

[yababom]

This really has nothing to do with the size of the Lionsgate catalog, and everything to do with the fact that there are no shortcuts to genuine experience.

The whole point of this attempt to use AI is a vain hope by the studios that they can take a shortcut straight to profits by leveraging the finished products of expert screenwriters, cinematographers, directors, actors, etc., rather than the actual experts themselves... So they feed the AI the finished movies, and hope it can spit out more finished movie

Why can't they make AI scenes?

[Somervillain]

You crawl before you can run. Why try for a movie when you can make a small scene? Here's why I think it's BS! We started with shitty weird computers then got decent, but expensive, calculators, then when I was a kid they were dirt cheap...we got to complex physics simulations by mastering basic arithmetic, then complex calculations. If I were an AI optimist, I wouldn't expect an LLM to be able to produce a full movie, but maybe a 30 second scene, of limited scope....master that, then move on to 60s, 5m