[slhomme]

As an owner of a Video Production studio, this kind of tech is blowing my mind and makes me equally excited and scared. I can see how we could incorporate such tools in our workflows, and at the same time I'm worried it'll be used to spam the internet with thousands and thousands of souless generated videos, making it even harder to look through the noise.

A fun related experiment, I thought it was fun to see what kind of movies AI would generate, so I created a "This Movie Does Not Exist" website[1] that auto generates fake movies (movie posters + synopsis). It basically uses GPT-3 to generate some story plots, and then uses that as a prompt (with in-between steps) for Stable Diffusion. Results may vary, but it definitely surprises sometimes with movies that look and sound amazing!

[1] This Movie Does Not Exist: https://thismoviedoesnotexist.org/

[skocznymroczny]

Reminds me of South Park when Cartman was pretending to be a robot and was made to invent movie prompts

"Adam Sandler is like, in love with some girl, but then it turns out that the girl is actually a Golden Retriever. Or something.""

[spaceman_2020]

“AWESOME-O”, S8 E5, if anyone wants to check it out.

A South Park classic imo

https://en.wikipedia.org/wiki/Awesom-O

[Dracophoenix]

Or the "Two Brothers" skit from Rick and Morty.

https://www.youtube.com/watch?v=ba9k5SWwE38

[raghava]

Once a capable AI that can make a "Two Brothers" trailer like this one below comes up, humanity has some serious issues to tackle, other than climate change ofc! :D

https://www.youtube.com/watch?v=tBUUoyZoja4

[maxov]

I love this! But after trying it a few times I got this result :). So fascinating.

https://thismoviedoesnotexist.org/movie/the-terminator

Brings up the age-old question of how much the learning in these models is just memorization. Though in cases like these it’s hard to tell.

[slhomme]

Yep, that's because GPT-3 was trained on real existing data, and it's quite a challenge to make sure the story plot is 100% fake. When it's too close from an existing film, it just sometimes gives it the same film title. I have in-between GPT-3 prompts to avoid that as much as possible, but sometimes real movie titles slips through the cracks. Something I hope to improve shortly.

[clipper_janosch]

What a great project, you absolute legend!

[sdenton4]

Given Hollywood's proclivity to remake everything on a 20-year cycle, it seems completely appropriate to get a 2023 Terminator reboot in a 1920's style.

[yazaddaruvala]

It also seems completely appropriate for The Terminator to be written, directed, and acted by an AI.

[gedy]

Here's "The Terminator by F.W. Murnau" from Stable Diffusion:

https://ibb.co/RN5bxJb

[WheelsAtLarge]

There are only so many stories to tell. The Terminator is a rehash of so many other previous stories. The real art is in putting it together so that it seems new and fresh and gets people exited about it. The Terminator 1920s style looks interesting.

[jonny_eh]

If a story is a rehash of "many" stories, then it's actually a new story. Similar to how an "Airbnb but for dog walkers" isn't actually a ripoff of Airbnb, but is in fact an original idea.

[acheron]

> The Terminator is a rehash of so many other previous stories.

I know what you mean, but I also laughed at that.

"I bet one legend that keeps recurring throughout history, in every culture, is the story of Popeye." - Jack Handey

[soperj]

Yeah, just got: https://thismoviedoesnotexist.org/movie/the-legend-of-zelda-...

It's crazy that it just made up those names...

[kurxig]

Since the generated output is so close to the training data, the model is probably overfitted and trained with too few data...

[trop]

Note to mention "In the Land of Oz: The Search for the Wizard":

https://thismoviedoesnotexist.org/movie/in-the-land-of-oz-th...

Which reads like a bad translation of a bad translation. Like the the old joke about the AI program which was supposed to translate "The spirit is willing but the flesh is weak" from English to Russian to English, and after the roundtrip came up with "The vodka is good but the meat is rotten."

[babyshake]

Not sure how I feel about an AI generating a movie concept that involves a "rise of the machines".

[maxov]

I feel like "generate" is kind of a strong word in this case though. At this rate if the machines rise up, they will do so just to parrot all the "machines rise up" plot synopses in their training corpus.

[pfalke]

Here’s another Terminator one: https://thismoviedoesnotexist.org/movie/terminator-war-of-th...

[firstcommentyo]

Common, this is genius. Plot reads: in 2154 a soldier from the future is send back in time to the present.

Why send a soldier from the present to the past when you can also have a soldier from the future send to the past!

[LambdaComplex]

I got this on my 4th try or so: https://thismoviedoesnotexist.org/movie/alice-in-wonderland

[anshumankmr]

The same could technically be said about Disney which is just remaking their entire classical collection but with CGI.

[hiidrew]

I think some refer to this as the dead internet theory, e.g. AI content creation becomes the majority of media on the internet instead of humans posting (may be wrong in my explanation but think that's the premise).

It's scary to think about it but seems plausible—like if someone can make an app with Tiktok-like ubiquity of only AI content. Although to your point I imagine there will be so much nonsensical noise that curating will become a useful skill, it is today but even more so.

[Dragonai]

> It's scary to think about it but seems plausible—like if someone can make an app with Tiktok-like ubiquity of only AI content. Although to your point I imagine there will be so much nonsensical noise that curating will become a useful skill, it is today but even more so.

This just gave me a disturbingly vivid vision of exactly what you described. A seemingly likely future where everyone's in their homes scrolling endlessly on a TikTok-like app where there's literally infinite content being generated by the AI all the time, and as people like and dislike certain types of content, the AI just gets better and better at generating new videos...this is honestly kind of terrifying. I have no doubt this will exist one day, and that it'll print money for one company while billions of people are spending all their free time consuming it.

[johnfn]

Well, what's so scary about this exactly? That's exactly what people do today on Tiktok, modulo the part where AI generates the content.

[wayeq]

plus the part where NPCs generate the content.

[Atheros]

People like interacting with other people, not bots. If a few individuals start to feel like they're interacting with too many bots, they'll retreat into small private silos. The human-to-bot ratio on public forums then drops. Then more people realize that they're just talking to bots and further and further it goes until practically no humans are left. That's the dead internet theory.

I really don't think it applies to us in this context though because I think that a decent number of humans don't care whether some content is AI generated. Furry porn is all hand-drawn and people still like it despite it not being real.

[moron4hire]

>> spam the internet with thousands and thousands of souless generated videos

Unfortunately, that's already happening.

https://www.youtube.com/watch?v=w7oiHtYCo0w

From what I can see, YouTube has done quite a bit of work to cleanup YouTube Kids, but it's kind of an arms race.

There's this worrying issue in AI ethics discussions where most people seem to assume the problems and dangers of AI are still off in the future, that as long as we don't have the malicious AGI of sci-fi stories, then AI and "lesser" algorithmically generated content isn't harming society.

I think that's not true at all. I think we've seen massive damage to social structures thanks to algorithmic feeds and generated content, already, for years now. I don't think, just because they aren't necessarily neural-network-based, doesn't make them something to not worry about.

So I don't see AI as a particularly, different, worrisome problem. It's an extension of an already existing, worrisome problem that most people have ignored beyond occasionally complaining about election results.

[agentwiggles]

This one is too funny: https://thismoviedoesnotexist.org/movie/the-virginity-pact

[BillSaysThis]

Why would this title bring in superheroes?!? lol

[pkulak]

Ooo, check out that trademark infringement.

[twobitshifter]

Saw this clip today in response to a tweet by musk saying a cyber truck can act as a boat. I thought either this was made by an AI or it will be soon.

https://twitter.com/dvorahfr/status/1575508907593711618?s=46...

[barbariangrunge]

I know some extremely hard working independent filmmakers who struggle so hard to get noticed. After this tech goes mainstream in 5 years and gets really good, I don’t know what they’re going to do

[elephanlemon]

Make blockbusters, because scenes that would be incredibly expensive to shoot now will be practically free with this?

[barbariangrunge]

Every 15 year old kid will be making their own blockbusters

[the_lonely_road]

Have you talked to the average 15 year old kid? Hell, have you talked to the average 50 year old? It will be the same as ever, a sea of absolute shit surrounding some true gems, be they from novel creativity or just excellent execution of well worn ground. The role of the trusted curator will rise and brands will gain more power.

I am sure I will watch MY 15 year old's attempts, and maybe a few from my extended circle but most of my consumed content will still come from what makes the cut to Netflix or HBO etc. Technology like this will empower the truly creatives once it has matured. I would expect closer to 20 years than 5 however.

[Nursie]

> It will be the same as ever, a sea of absolute shit surrounding some true gems

Ah, I can see what's wrong there, just turn down shit randomness and increase the true gem sampling steps to proportionally increase the input weight of true gems and fix your output quality.

Basically - if it empowers creative people, their output will be fed back in and parameterised.

[gpderetta]

Better, you ask your local AI agent to filter according to your personal preferences and general artistic qualities :)

[have_faith]

From my personal perspective; knowing how something is made affects my interest in it. I guess that's why provenance ascribes value. I value human creativity and so I'll likely always seek out something created by craft than by shortcuts.

[electrondood]

AI is eating the world, and the vast majority of people are not paying attention.

I don't know what artists, truck drivers, Uber/Lyft/taxi drivers, delivery drivers, programmers, doctors, judges, fast food workers, etc. are going to do.

[ddalex]

Perhaps humanity would happily share the benefits of machine work and we can all spend idyllics lifes eating, laughing and loving while exploring the galaxy?

[jmfldn]

Under a different economic system maybe. Have you seen how our one actually works? See the film Elysium for a more plausible future.

[LambdaComplex]

Great idea!

Under the current system, the rich can do just that while everyone else literally starves to death :)

[astrange]

If they’re not trading with everyone else they’re not rich. Being rich /is/ the ability to trade a lot.

The idea that rich people will all leave and start a different rich-people-only economy that somehow takes all the economic activity with it isn’t how it really works, it’s the plot of Atlas Shrugged.

[pkaye]

Then the rich will have no consumers.

[ericd]

Why do you need consumers when you have machines making most of what you want? Then you just need to be able to trade things with other people who have different machines and resources from yourself to get the bits you can't have made by your machines.

[rcarr]

Maybe they’re after some Highlander style planet where « there can be only one »

[EwanG]

So "Childhood's End" without the aliens?

[sh4rks]

> programmers

Surely if we get to the point where programmers are no longer needed, humans have essentially been replaced by AI? Since the AI programmers could just program better and better AI?

[munchler]

Yes, this is called “the singularity” in AI circles. Personally, I don’t think it’s imminent, but it’s certainly worth some concern.

https://en.wikipedia.org/wiki/Technological_singularity

[spaceman_2020]

Really have a feeling that if you don’t build your nest egg in the next 5-10 years, its going to become incredibly hard.

[rnk]

Why will conventional investments have value? Why will it be safe? I could be retired and well off with say even 5 or 10 million and then large numbers of people lose their jobs, yet we can still make widgets and grow and distribute food with robot everything, society will fall apart. There will be fewer people with money to support themselves or buy things. I don't see the capitalist society in the US giving people basic income, but we won't need so many workers, and so anarchy.

[codegeek]

doctors ? We are a looooooong way to go before AI can replace actual doctors.

[dubitably]

A month ago I would have said that about actual artists.

[extragood]

Sleeping Beauty actually looks really good: https://thismoviedoesnotexist.org/movie/sleeping-beauty

Edit:

Worst (best?) tone clash: https://thismoviedoesnotexist.org/movie/stalked-by-a-friend

[agentwiggles]

That second one is seriously hysterical

[usefulcat]

> I'm worried it'll be used to spam the internet with thousands and thousands of souless generated videos

I agree, and I think when that happens, it will tend to increase the value of curation. High quality curation that is, probably done mostly by hand, as opposed to the at-best-mediocre automated curation that is commonly used.

It could be bad for things like YouTube, for example. I think there will be an arms race between generated video content one one side and automated curation on the other. I mean, you can still leverage viewer choices for curation (looking at what people are watching a lot of), but that is just shifting the burden of curation to users. Few people will be willing to sift through dozens of cheaply generated crap videos to find something they actually want to watch.

[ericd]

The volume will increase so much that the only choice will be automated curation tools, at least as a first pass. The arms race is on.

[cutups]

I was thinking about something like this site, but also taking some randomized existing plot outlines to generate specific stills from each part of the plot. Might require isolating character archetypes too?

Great work with this as is!

[pmarreck]

https://thismoviedoesnotexist.org/movie/the-yetis-last-stand

The description text does not convert sentences with carriage returns (or probably, newlines) into separate div's or whatever html element you'd prefer, FYI! Otherwise, very cool!

[WheelsAtLarge]

Cool toy. One of the most useful side effects of AI right now is idea generation. Market this as an idea generator for movies and such and people will eat it up. Try posting it on the entertainment focused area of Twitter and people will go nuts for it.

[klondike_klive]

These prompts are way better than the drivel I see on Amazon Prime. Half their movie descriptions don't even tell you anything about the film, they seem to be just a random paragraph from the pitch document.

[gcanyon]

"Soulless" -- they'll be low quality, both in rendering and in plot/acting, but they'll be anything but soulless. Each will be a labor of love of someone with a dream.

[sgrove]

You should add a “tweet this movie” button that pre-populates the image and the title! I immediately wanted to share one of the funny suggestions.

[sixQuarks]

For those who are scared about this technology, it’s good to look at what AI has done to Chess.

The best chess seems to be when AI is used along with humans. I think image and video AI will best be exploited when human input is also taken into account.

There is still something special about human creativity, I think AI will just be another tool to expand that. At least, in the short term I would say 10 years perhaps. AI will probably one day take over all aspects of creativity and humans won’t be able to contribute.

[dwohnitmok]

> The best chess seems to be when AI is used along with humans.

I don't think this is true anymore. I don't think I've heard about successful centaur chess games in years. I would love to be wrong there though (in particular if anyone knows about how correspondence chess games have been played in the last 2 or 3 years with the availability of Leela Zero and Stockfish NNUE).

[dwohnitmok]

Found this interesting article http://chessforallages.blogspot.com/2021/02/correspondence-c... which linked to this thread on the topic: https://talkchess.com/forum3/viewtopic.php?f=2&t=76382

Based on that thread, it looks like centaur chess is close to dead.

> Human input in top-level ICCF [correspondence games with chess engine support] games is now 99% eliminated, other than personal preference in selection of openings.

[dotsam]

Don't worry, humans will still be relevant for ~10 years!.

(and then regrettably irrelevant thereafter).

I think it is a legitimate worry, as the pace of progress is considerable. These tools are impressive, and are only going to get more impressive: more people should be talking about where this is headed.

[trention]

These models are literally one French court decision away from being banned and investment in them being completely halted. So I wouldn't worry too much about them specifically.

[tehsauce]

I think the opposite is true. I think we're closer to the beginning of a kind of arms race.

[zepolen]

Yes because banning alcohol, drugs, porn, prostitution and profanity in music sure got rid of them!

[trention]

It would have gotten rid of every single one of those if the only method for their production required a tech company to invest a 8-digit sum in R&D and said tech company was threatened with litigation.

[Nevermark]

Porn companies, Facebook, Uber, ... these are just the latest companies, in a long line, for whom huge lawsuits and fortune-sized settlements are simply a line item on the expense sheet.

[rnk]

That is nonsensical. Maybe one country might decide to try to stop it, the internet has shown that doesn't work.

[ronsor]

Why French court? If other countries don't ban them, France will just be hurting itself.

[trention]

A french court banning it will probably be a (short) precursor to an EU-wide ban which is effectively a worldwide ban. EU courts will be fining companies heavily if they don't comply which effectively means big tech no longer invests a penny in models. I use 'French court' here mostly because I believe it most likely to be firstly litigated there.

[CalRobert]

I imagine plenty of companies in the rest of the world will happily make use of it.

[rnk]

Yeah, just like Facebook and Google can't operate in Europe, or the US isn't still spying on the whole world still. Those eu courts really control them.

[abeppu]

I think a key difference here is that with chess, 'goodness' is defined by winning. With content generation, the training methods point towards some form of comparing the generated thing to some observed data, but the 'goodness' of the content from the perspective of potentially competing with or displacing human creators is "do people like to consume it?"

If one trained using e.g. a tiktok like dataset showing viewer response measurements for each video, and do conditional generation on those response values ("prank video watchers are highly likely to watch the full video"), are we really that far from a system that learns to generate content that attracts and hold eyeballs? Not so long ago there were a lot of concerning trend pieces about how youtube had a network of creators making bizarre, disturbing or transfixing videos being watched entirely by young children. Before that, it was clickbait listicles. "Bad" content that can get eyeballs can still wildly steer what humans create and consume. I'm wondering if in 2 years we'll have an enormous number of short videos that we all agree are "bad" but which are nevertheless constantly watched.

[looknee]

I may be mistaken but I believe that human/machine pairing was dominant for a long while, but the last few years the chess solvers have progressed to a point where they're dominant on their own.

Poker on the other hand I think human players still win vs GTO solvers, but again I may be mistaken here too.

[lairv]

I think AI now beat humans at poker

https://www.sciencedaily.com/releases/2019/07/190711141343.h...

https://www.cs.cmu.edu/~noamb/papers/19-Science-Superhuman.p...

[LarsDu88]

Is there a single "intellectual" game left that humans can beat computers at? I suppose an AI has yet to beat lebron james at basketball, but I suppose that's for want of having a body.

[dwohnitmok]

> Poker on the other hand I think human players still win vs GTO solvers, but again I may be mistaken here too.

Also an outsider, but I think this has changed in the last year and that AI now is consistently better than top-tier humans at even no-limits poker.

[astrange]

What does “best” mean here?

AI is the winningest in chess, but the real life purpose of chess is to produce interesting gameplay for people to watch, and so AI is less good than Magnus at that. You’d need the AI to throw games and write press releases.

[nonima]

>it’s good to look at what AI has done to Chess It completely ruined the game to the point where it's more about memorization than it's ever been.

[synu]

There are two classes of engines. One is like you describe, faster and faster brute forcing. AlphaZero was much more creative and didn’t use brute force.

[fooblaster]

top tier chess ais crush human grandmasters and achieve super human performance with no assistance

[wayeq]

what? Alphazero trained via self play and in ~10 hours became unbeatable by humans

[hugozap]

Is the generation happening in real time? I'm curious about the costs of running something like this.

[slhomme]

Yes it's happening real time, so far it's been generated about 8k movies over the last 3 hours. The costs right now are roughly about $150 for the generated images (Stable Diffusion) and $30 for the generated texts (GPT-3).

[Yajirobe]

$180 in 3 hours?

[_tom_]

First one I got was "the time traveler's wife" which does exist.

[samstave]

Ha - getting app too busy errors so can’t see your site…

[slhomme]

Sorry about this, we're having a hard time handling the heavy traffic coming from Hacker News. Over 4k movies generated in just the last 2hours, this definitely impacts our servers performances, not mentioning our Stable Diffusion and GPT bills! hehe. Currently working to make things smoother!

[samstave]

This is normal, so appreciate growth