How Kpopalypse determines the use of AI-generated imagery in k-pop music videos

4 days ago 6

Caonimas have been noticing that AI-generated visuals have been appearing in k-pop music videos more and more over the course of the past year. At the same time, there’s also been a rise in the denial of the use of AI in music videos by k-pop fans eager to cape for their biases. But what is the truth, and how is Kpopalypse determining the use of AI? This post has the trufax!

When AI-generated imagery started appearing in k-pop music videos, I originally wanted to boycott all use of AI imagery everywhere in roundup and Kpopalypse coverage in general. My stance on this was because of the following reasons:

AI imagery takes away jobs from artists
AI imagery is basically plagiarism
AI imagery isn’t very environment
AI imagery generally looks like asspaper

Unfortunately for me, this moral high-ground proved to be unworkable. The use of AI has now taken off to such an extent that if I stood true to that boycott I’d probably be left with a large chunk of the most interesting comebacks not even covered, and there goes half the stuff in roundup people actually want to see. So the only stuff that I’m boycotting is the super lazy stuff, the videos that are literally just an AI prompt into a video-maker like Sora and not much else. Please now thank one of my readers as I introduce you to AI Jesus:

This terrifying video showcases typical lazy AI video generation results and shows a lot of the things that AI can’t do right consistently yet, like draw the Roman numerals on clocks correctly, figure how how many digits people have on their hands, or stan Chuu in a sensible fashion.

Of course it’s easy to pick on an independent artist’s effort, and as we all know they have no money so we can probably forgive them (a little), the temptation to use AI instead of hire another starving artist (like them) must be large. Any k-pop idol studio wouldn’t have those kind of worries though, they’d tidy up such little details before the final product and make sure that no mistakes like that slipped through anywhere because as we all know k-pop is all about consummate professionalism and creating the best product possible…

Well, okay never mind. But my point is that there’s no real debate about these kinds of uses of AI, it’s fairly obvious. K-pop fans definitely noticed that the above video had lots of AI in it, to the point where the label sat up, took notice, and released a “natural” version with all the more obviously AI-generated imagery removed but plenty of the less obvious stuff still left in no AI-generated imagery at all, gosh no, pinky swear.

This of course completely satisfied the concerned k-pop fans. Isn’t it lovely when agencies listen to their fans, gosh yes. However sometimes it’s not so easy to determine if AI was used. So how can we be sure?

Whenever I determine the possibility of AI use in k-pop music videos, I try to look at it from a business perspective first as if I was the creator:

What would be the time/money investment for AI to do this
What would be the time/money investment for a human to do this via physical props, CGI, hand-painting etc
Where am I actually going to use the result, how much bang for my buck do I actually need given the intended use

An important thing to understand about video directors choosing to use AI, is that they’re typically not “AI-bros”. They’re not like the mouth-breathing idiots who hang around on social media with NFT profile pictures talking about some techno-optimist bullcrap, they don’t have any ideology invested in AI use. K-pop video directors are simply using it because they have production deadlines to meet, and they want to use the quickest way of doing something, like a programmer who might use AI to knock up a quick code template just because it saves a bit of time. That doesn’t mean they’ll use AI for everything, definitely not, because 1. the tools suck too much to trust them to that extent (see AI Jesus above) and 2. why automate yourself out of your own job, but something conceptually straightforward that’s easy enough to explain to a computer with a prompt but a tremendous hassle to actually make manually is something that a k-pop video director will be very tempted to use AI-generated images for.

So let’s look at the main types of AI use in k-pop videos today, by singling out some videos that I’ve previously flagged as having AI use, and look at why AI might have been chosen in these types of circumstances.

Hyuna sorry I mean IU’s “Holssi” has a video which is mainly not AI, but the floating people certainly are AI. Indistinct-looking people with withery, undefined limbs, hands and feet is definitely something that’s directly within AI’s wheelhouse at the time this video was made. This type of look is something that AI image generation can knock up quickly, but which would be a true pain in the ass for a human to do with a manual CGI tool, so it makes sense that they would use AI here for this. It’s probably also the reason why this song has a “recording behind the scenes” video but no “on the video filming set behind the scenes” video like IU usually has for all her other MVs. Why would IU’s agency miss an opportunity to do a behind-the-scenes video for a big inflatable Tweety Bird… unless they didn’t want people to know that it was just a “big yellow cartoon bird please, plagiarise Warner Bros a bit why not fuck it” prompt?

The dog/wolf/whatever the fuck that is at the start of Kiss Of Life’s “Get Loud”, that’s AI-generated for sure – no, not CGI. How do I know?

CGI dogs these days look a fucking hell of a lot better than this shit, look at its feet holy fucking christ where’s AI Jesus when you need him
Who the fuck has time to draw a CGI model for a dog that barely gets used

The second point is crucial. If it were a manually-built CGI dog, they’d probably make it the main feature of the goddamn video and use it a ton, because it would have taken some artists quite a bit of time and effort to actually create it and they’d be thinking “now we’ve fucking paid for and got the damn dog, let’s make the most out of it so we can get some return on our fucking cyberdog investment”. However, they didn’t do that – they’re only using the dog for two seconds at the start of the video, so what do you think is more likely for them to do:

pay someone for all that time and make them put in all that effort to (badly) animate two seconds of dog that nobody cares about, or
someone in post production thought “gosh this scene is featureless and bland hey a dog would look good here” and threw a prompt into some AI and made a dog real quick to fill up the space and threw it in and the director looked at it said “has that thing got paws of hooves who the fuck knows but hey fuck it k-pop fans will lap up anything as long as their bias is in it, I guess it mrds”

There’s lots of floaty AI-generated crap in Odd Youth’s “Best Friendz” video, like random bubbles, confetti, and… people having accidents, how aegyo, much heart shape. Again, it’s not manual CGI because think about the options:

Pay a stuntman to fall off a ladder, multiple times
Hand-CGI draw in a realistic-ish looking guy falling off a ladder
Type “workman falling off a ladder” into some prompt somewhere

It’s only background detail (and it looks like dogshit anyway, notice how all the angles are deliberately so you can’t see the face, hands or feet, plus where the hell are they landing) so you know they are going to go with the laziest possible option. Again if it was someone’s blood, sweat and CGI tears the falling would be more front and center, it wouldn’t be a blink-and-you’ll miss-it background detail. There isn’t time to sweat over background details when there’s production deadlines to meet. AI for sure, 100%.

There’s also a technique in AI image generation that I like to call “detail spam”. Watch the sequence of images in Achii’s “Fly” video from 2:30 to 2:36. This is all AI-generation at work. Whenever you see something in a music video that fits the following criteria:

Easy to describe to a machine-learning program with a prompt
Extroadinarily high amounts of rendering detail, which would take a real artist hours or maybe days to create
On the screen for an absolute split-second

Then it’s AI-generated (because again, if it took a real human days to make it, you’d better believe you’d see it in the video for longer). You see AI used like this a great deal in k-pop videos for transitional effects now, that flash up very quickly. When working with traditional CGI, a transition with a ton of detail is a headfuck, but with AI video generation, a heavily detailed transition isn’t all that much harder to do than a simple transition, because you describe it to the machine using the same method.

Same again with Jay “where’s my soju” Park and “Gimme A Minute (to type in this prompt for exploding cars)”. We know this isn’t hand-made CGI because hand-made CGI cars actually look really fucking good these days, CGI artists worldwide have honed their craft on decades of shithouse Vin Diesel movies, they know how to make CGI cars look like actual cars, more or less. The cars here on the other hand suck donkey dicks, they have the same weightless, blobby, wishy-washy feel as the floating people in the IU and Odd Youth videos, and they look like absolute ass, which is why they’re on the screen for such a short amount of time, so you don’t notice how badly a job they’ve done.

AI use doesn’t have to suck in the end result though. Feel however you want about the ethics of it, the right creative minds can still get decent results from it.

XG probably represent (at times) the pinnacle of AI-generated imagery in k-pop at this moment (and I’m calling XG k-pop for simplicity’s sake because idgaf). Plenty of XG fans were quick to insist that AI wasn’t used here, because why wouldn’t they, they gotta cape for their bias after all, using shaky arguments like “there’s behind the scenes photos of props plus a big credits list of people who worked on it, therefore there’s no AI“. Total bullshit of course, it doesn’t matter how many people worked on it, or how much of it is manually-built props, they could still be using AI as well. XG use AI in their imagery all the time. For an example, check out the “Princess Mononoke”-inspired foot imagery at 1:20 in the video. Fans were insistent this wasn’t AI because this photo exists:

It’s a legit impressive prop for sure, but it doesn’t mean AI wasn’t also used. This is actually precisely how the best visual designers work with AI (and CGI as well), they do as much stuff with manual props as they possibly can (because it looks better) and then add in CGI or AI-generated images afterwards over the top to enhance it even further (in this case the ‘growing’ effect on the ground).

It’s highly likely that XG used AI for this and other transitional effects in the “Howling” video, because… well, they’ve done it before, so why would they stop now? Their credits for the “IYKYK” video have four AI artists credited. I promise you that they didn’t turn around and suddenly say “gosh AI isn’t very environment, we’d better stop using AI artists now to help our country reach its carbon emissions targets or the fans will have a sad”. Sure, everyone cares at least a little about the environment, but when there’s a production deadline to meet, those concerns tend to become someone else’s problem. At least Cocona’s shaved head is real and eco-friendly.

Speaing of all things environment, I’ll leave you with environmental expert Chuu’s “Strawberry Rush” which is almost certainly using a fair bit of AI-generated imagery for all the more boilerplate-looking background cartoon shit. It’s not the first time Chuu’s new agency have used AI and they’ll probably keep doing it because even though the fans hate it, as long as they keep their interest up the agency doesn’t care. It works okay in this context anyway because it fits the surreal nature of what’s going on, and there’s a good chance that the AI output was then curated by humans and modified a fair bit further. I suspect that the main thematic details here were manually made and then some AI stuff was added to fill in the gaps. The result is a video that looks good but probably a bit too crowded for its own good, it’s a bit like watching those remastered Star Wars films where George Lucas added all the extra creatures just because he could. Also it’s hard to see all that cool Melbourne graffiti because of the critters getting in the way, it’s just as well that Chuu has a big enough screen presence to cut through it all.