I regret not making an Arctic monkeys joke. Remember to check out the description to pick up the Essential Topology Guide for 40% off until the end of November.
no, its because the AI that exists to do these videos only can render small segments into motion. I've seen a bunch of clips of how to do these with various software and its basically just taking a still image and animating a couple of seconds of motion into it. These AI clips are run by prompts. You have to be very specific just to get an image that is reasonable. And it can only extrapolate a very small amount from that point on before it gets beyond the prompts and starts to get really wonky. As is they probably had to have the AI animate each of these images a thousand times just to get something this good.
As long as I know the art is AI generated then I know it is bad but if I can't tell then it won't be bad or good until I know if it is ai art or human art
No. I mean yes. It can a bunch of reasons. The model used might begin to quickly fall apart and introduce to much crap wrong, very quickly. Some of the shots might in fact be 3d scenes rendered and then taken a single frame to let the AI sauce create a believable 30sec moving shot out of it. (serious.) The actual main reason for it? Trying to make a advertisement with this tools is like trying to paint a painting with a mouse cursor. And no, I do not meant in Paint. But trying to make the screen cursor go out of the screen and pick up the pencil and paint on the canvas in the room. Generating still and moving pictures with AI? In the now 'traditional' meaning of it? You type text into a prompt. Like this very comment is done. But this comment is not trying to recreate a cinema grade advertisement for cola. What would a machine-learning prompt look like? Have no real clue how a cinema grade advertisement for cola would completely look like. But can give a very short and badly done one: Red truck driving on a winter road. Time of day night, snowy, calm, sooding, $$$$$, imput emoj here. Try and imagine crafting a text prompt that can be used! You have to write up a strange machine-learning oriented scrip for every scene. Shape and coerced the input prompt until you get something usable. And repeat this endless cycle of getting the machine-learning to produce a scene that is what your after. This is not magic. Anyone having done 3d modeling or anything like painting/shooting movies/pretty much anything requiring a skill to make something look realistic? We all know that it is all a big amount of fakery. Polygons and paint strokes. Even if depicting a real event and real objects? It is only that. A depiction by means of something else. Machine-learning being marketed as AI is simply that. The amount of control some machine-learning creative code can do is simply limited to what it has learned. And the control we have over it is rather dull. If a pen or paint brush is not working as you want it? Change it. If a 3d program is not behaving as you want it? Then with source code change it. If a AI Machine whatever is not behaving as you want it? Then good luck getting it to behave. With almost no re usability. You have to regenerate everything for every scene and change. And I mean not re-render the lighting. But every pixel and so shape of the hole world is going to change. This AI machine whatever codes can not even be relied to make trucks have a look coherent to each scene. You can probably try and tell it to make: American, Truck, 18wheeler. And yet you are not really in charge here. And instead get some European truck. You can try and describe and input as much as you want. It simply is not going to do it reliably. It is so out of the control of the 'artist' as you can get it. Simply to much is up to randomness in the code. Machine-learned-code. Gibberish. We need actual AI to understand 'AI' code. Meanwhile a photograph or voice recording from 1700s can be put into a scene or used as is, anywhere. Edited etc, etc. Converted to digital or analog back and forwards. A machine-learning prompt might be good to try on a different model? Plain readable text. Not actually something worth alone to be reused. Not even sure if you can make a machine learning model recreate the same output twice. At least when it comes to this scale of machine learning. Quite allot of things need to be made up on the spot to create something out of tin air. And it is out of tin air. It is using what is learned. But how do you pick what to use of what you have learned? Creativity. Intuition. And that is exactly what AI lacks. At least while also keeping track of everything. One can not go into a AI-generated scene and change something. Since if you try? You are going to regenerate everything and something completely different is going to exist the next time you hit render. At best you can feed the AI with AI created things? And hope that it dose not learn from AI mistakes. AI is pretty much perfect for Photoshop and such effects. We have had AI for a long time. We only called it Photoshop. Trying to use Photoshop to make a Cola advertisement alone? You would pretty much make a animated moving image. As is trying to use 'AI' to make a moving image. You get the results based on that cause. Cause and effect. Text prompt and weighed values? That is what the 'creative' person get to actually control. Then it is up to the AI to make something out of that. Try and use Photoshop with a text prompt. That is what this is. With no ability to go backwards in the processes. You get a single text prompt and Photoshop makes a hole scene out of that. You have literally no control. Only hopes of something good. You might as well outsource the work to someone else at that point. Being a artist. How many, and how much time was used to generate this scenes? It is very likely a embarrassing number... And if not for the panning camera? The scenes are rather static. Moving heads. Moving trucks. Very short shots. All because it can not be controlled. You could literately write a machine-learning prompt and send it to someone that do this kinds of projects for a living 'the old way'. And you would actually get a better output. Since I do not know, maybe using Photoshop alone is a really bad idea when making a short movie/advertisement meant to look real or good in any way? Use the lasso tool to make a hole movie! Maybe I'm wrong. Maybe this AI Cola add was from begging to end generated. Without editing of any kind. And the AI also had the pleasure to export the finished file and upload it to... wherever ads go. Or this are a bunch of random scenes stitched together. Hardly possible to make a single 30sec scene look good enough to fool the general public to not see it as a bubble. If I was to draw or model/animate a reindeer? Or really anything? I would look up and find images and such to make it look right. I wounder what data machine-learning uses. It can not be looooots of data to do what a artist would be able to do as well without image references. Almost like humans are better then a machine at learning. And without the need of braking laws.
As someone who works in an arthouse this has the fingerprints of junior artist's first after effects job all over it. All the branding are just PNGs tracked in, hell, all the coke bottles are just PNGs. All the lights turning on are simple masked altered footage and you can spot the comped shots clear as day; they're egregious. Coca Cola cheaped out on this, big time.
It's incredibly apparent that nothing moves in three dimensional space, but rather in the style of those animations that convert hand drawn pieces of art into motion.
Ai animations are suffering from being blurry and also objects with ai animations never actually move, they almost always seem to morph around the screen
"Cars can only go two miles an hour so my horse is sure to win" It's over, bro. The 3D industry is finished and people are in varying stages of denial over it. The last of which is criticizing the "art" of commercials for corporate clients.
@@jeanclaudethedarklord6205 When's the last time you paid a team of people to do in six months what you can get in ten seconds for free? The 3D industry is done. Commercials, movies, games, television shows, art commissions - already being outsourced to AI after being outsourced overseas. People do not notice a difference and they do not care.
It's already in effect. AI generated channels with AI / bots generated comments, and artificially inflated / bought for subs. Here on youtube it's already all over the place.
Yeah I've seen them. I just didn't think this was supposed to be hot water. Not enough steam. I'm still not sure what monkeys have to do with Christmas
Agree, it looks like cold water but they seem to sill have fun😊 also japan is definitely not famous for celebrating Christmas. I think they just did all that with AI for getting attention and it works
I find that using AI in advertising like this accomplishes the opposite goal because the more times it’s shoved in your face, the more you notice the weirdness and it becomes more unappealing. AI stuff works best when you only see it once in passing not over and over and over like Modern advertising
As a VFX guy myself, a big company like Coca-Cola should be very careful doing an AI commercial like this, because AI sources the imagery from multiple sources. The risk here is some source elements getting mixed into the advert, that has come from an artist or company, that Coca-Cola haven't had permission to use. Could result in them being sued very heavily. AI imagery and video still has a lot of question marks over the head, and with this new tool soon coming out, that supposedly will be able to unscramble the AI and source the elements. When that happens things are going to change massively.
Well, the actual risk is infinitesimal; I've made an experiment with MidJourney, prompting it to recreate some famous paintings. With Mona Lisa it was true to every detail, even in the background. Picasso's Guernica was recognisable, but various details were missing and patched up inconsistently. Munch's Scream had the overall composition, but the screaming character was created completly anew, more realistic and not in the Munch's style at all. As you can imagine, it went downhill from there. It couldn't recreate even the basic composition of Rembrandt's Night Watch, it returned a generic period scene in the "style" of Rembrandt. So unless something is as iconic and overrepresented in the sample base as Mona Lisa, the AI is not likely to recreate by accident any details of any single piece of art sampled in there.
Some of the animal shots apparently John Carpenter originally deemed too cute for the theatrical cut of The Thing, and they eventually ended up here. An other testament to the power of film, scanned in beautiful 4K about 40 years later.
I don’t like AI ads, but it is crazy how far AI has come with like stablediffusion. Just think how far it’s come in just like 2 years, from the will smith eating spaghetti ai video to this.
This is a totally different AI coke ad than I've seen people mocking on social media, the one I've seen has coke bottles changing size and shape in people's hands.
@5:39 - You can see a tracking error with the logo on the second truck - definitely looks added in. @8:13 - The back of the second to last truck has a messed up logo. Probably one they forgot to replace with a "clean" one.
Probably. I churched out about 30 different AI thumbnails for this video, and it messed up the coke logo almost every time. And that was using Flux, which is one of the best algorithms that's particularly praised by AI bros for its skill with text. 🤷♂️
The monkeys are Japanese macaques, the only monkeys native to the archipelago, who are know during winter to enjoy spending much of their time in hot springs. The ad may have a lot of problems, many of which you pointed out, but showing the macaques in the hot springs just after the trucks passing trough a Japanese city is a very intentional human choice derived from reality, and not one driven by AI madness. Also, side note, japan loves Christmas and they even have their own unique traditions built on top of the American/European christian ones.
It always makes me perplexed. These ads used to be so pixelf***ed for accuracy and perfection. But as soon as AI bros got in there.. it's like all kinds of quality control went out of the window. So many of these issues could probably have been corrected by simple compositing and paint overs or just regeneration. But the only postwork here seems to have been the logos...
You are being so nice with this shots You say "This bunny, I supose it's ok", but it's fur is completely fckd up "This trucks are not bad", but one truck is emerging from another one! And so and so lol
This is spit in all our faces! How many hours did we all spend because a water drop did not seem realistic to the client and they threatened not to pay us, unless we make that water drop look more real than real. Or calling us out if we could not match a multi million dollar production, for a few hundred bucks they are willing to pay. I wish someone would put Cocaine back into Coke. That would make one hell of a commercial!
I'm pretty sure all the lights were burning bright in each generated shot, and they manually dimmed them down at the beginning to create the "turning on" effect.
I just hate how big companies just shuts their eyes to AI's flaws but if it was made by some videoprapher or CGI artist they would be pin pointing the tiniest thing to make it perfect. And let's agree it always looks unsharpened, blurry and low resolution.
I hate it too but that makes complete sense. If a single artist remade this entire ad, then first of all it would take 200 times longer to make. And since humans are supposed to be better than AI (that's what most artists seem to think), then companies will expect little to no mistakes. An ad like this, only made by humans would require a lot of resources to make perfect and respect deadlines. AI can do that for essentially free, and the result is decent, mistakes are tolerated. Give it a few years, it will remain cheap but the result will be perfect. It will still require human input for artistic direction, but much less than required before. AI video is just a few years old. The first generation is never the best.
I get what you're saying, but I think they should wait this few more years before it's perfect, then. And this kind of work are rarely done by a single artists but done by a hole studio. Those are HUGE companies with a lot of money. Hiring a studio would not be a problem and would result in a better result. But we need to understand the context here. This is a partnership of Coca-Cola with Dall-e. So it's nothing but business here. The money they're making with this partnership may be worth the poor quality.
I laughed my ass off at the mutant animals. The truth is 99% of the public will not notice a thing because grandmas will see a bizarre ai image of some kid in africa building a space ship and they will think it's legit, AND they get free marketing from the backlash for using AI. I think it worked pretty well.
"They went for the AI thing because they know it will get headlines." Well... The effect on me and my friends was pretty great. Now I refuse to buy Coca Cola and I showed it to my friends that normally don't even care about Coca Cola Christmas ads and now they also boycott Coca Cola.
You mention how the logo is always crisp and clean - but if you look at the shot with the driver, the logo is out of focus in the background and is, really, the only time it looks odd, especially the O on the word 'Coca' which is clearly independent, but in the out of focus shot, is merging with the c next to it!
AI is just a relatively new-tech tool which is going to get better and better in the future. While I like the MESSAGE of Coca-Cola about united people as a family in sort of magic and prosperous days, kindness and stuff, I dislike the high level of imperfections in todays AI. But I'm aware that could and will be fixed gradually, so it really does not bother me that much
I don't know if the first bear in the tree is meant to be a polar bear. The head shape is wrong, and while it may not be intentional (because AI), it looked tome more like a Kermode bear, which is a white/off white subspecies of black bear.
This reminds me of the amount of work required to make morph and warp effects convincing at post houses in the late 90s. I'm guessing - beyond all the work trying to prompt this amount of consistent-ish footage out of an AI - the ad agency artists and producer took ages checking, tweaking and trying to stitch together this ad from the mountain of generated, mutating footage..
The deers' necks get longer and longer. The hand holding the bottle has 6 fingers. All that is creepy. There are monkeys in Japan that actually do use hot springs.
Also It must have taken many interactions to get this sloppy result 😂 they could have done it in CG in the same amount of time using a decent asset library
I believe there was actually a lot of post on this. I believe all logos were tracked in, bottles were pretty obvious. I think some of the shots were comped together as well like the ground popping in when he puts the coke case down its actually just a crappy roto mask that wasnt feathered to get the contact shadow in.
@@error.418 it is not good enough for you and me, but it's good enough for millions of people with... "poor pattern recognition skills" (I'm trying to be very polite here).
@@pawe9082 yeah, and fast food is "good enough" for millions of people, but I feel like we should work to do better for our fellow humans. if we keep letting garbage be the most accessible thing pumped into the system, we just end up with more garbage, and that affects all of us.
@@error.418I completely agree. Unfortunately the big tech still has an aura of cool young dudes with noble mission. They should be treated like any other industry. I've never seen, for example, Exon Mobile CEO giving lessons to enthusiastic crowds about moral values, bright future for entire planet and similar nonsense.
I auspect the de-saturated lights "turning on" is not made by the AI. It probably created the whole shot with all lights on, then artists went in and manually de-saturated them in order to create the "turning on" effect.
In the grand canyon sunrise the shadows point about 45 degrees to the left of the sunrise. So, a second far brighter sun behind the camera? That sounds ominous.
the scariest side of the coke ai ads... it worked... everyone is talking about it. The vallée dérangeante (don't know the word in english) is still there and that's probably what they were looking for, that's a really disturbing idea to consider a world where robots are better seller than human. probably something who can generate a "small alignement problem".
8:10 you dont catch that the front truck has 2 trailers but no wheels underneath at the join. I dont know if trucks can have doubled up trailers like that, but I would imagine they would have some connection to the ground so the truck doesnt jackknife on every corner. also the back trailer has like 6 wheels as if it was a hotwheels toy :'D
OK, well I'm not a vfx artist (maybe one day, hence my sub) but to me, that looked like a big, shiny late-capitalism convoy through uncanny valley. Something unsettling about it, but I didn't really figure out what until you picked it apart.
I know we're all supposed to hate on this, but honestly I'm surprised it says good as it is. It has a consistent visual style throughout, the panning and tracking look like typical camera movements in video, and the continuity errors don't really jump out. I am curious how they trained it though, like what if someone had to essentially build the whole thing with a mix of real footage and cg, essentially doing the whole commercial traditionally, and then they had to recreate all that work with AI, using something like a style transfer to get the desired Norman Rockwell look, just to say that they used ai? I'm sure people initially are thinking 'why would they want to say they're using ai? Everyone hates it'... but I can imagine a possibility were there marketing team convinces them they should do something controversial on purpose, because it will get people thinking about coke, and realistically those people aren't going to refuse to drink it just because of General AI bashing. How many of us are currently thinking about or talking about pepsi? Somehow Coke has gotten me to see this commercial twice and I don't have a TV and I have ad blockers up the wazoo. Edit - oh, I should have waited until the end of the video, he already said it
The monkeys are totally relevant, the US is a little strange because we have the story with the reindeer, but in most countries outside of the US it's 12 Monkeys that guide Santa's sleigh. And instead of one being an outcast because of his red nose, it's the one baboon who snuck into the group who is shunned because of his red ass.
The reindeer at the start, had parts of its antlers just appear to detach and float in the air before disappearing (thought it was a tree at first but pretty sure it isn’t) The shot of the trucks after the monkeys, the front second truck seems to be merged into the rear of the font one. Also it seems to have lost all of its rear wheels
Missed in the back ground while the poor ducks' legs helicopter beneath them: the 2nd truck undergoes some kind of mitosis to become two separate trucks.
When one of the biggest companies in the world pulls this crap on the biggest marketing campaign of the year, it is nothing more than a huge middle finger to production houses and artists. To Hell with Cocky-Cola. We should just wiggle our bare arses at them and buy someone else's product. If they don't lose sales over this, everyone will follow suit, and it will be like bullet-time all over again. Thank you for doing this, because I wasn't going to watch their official video and give them the satisfaction of another view.
Yeah, tons of messy stuff in there, and yet they let it pass. This stuff would never fly in a TV show or film. I think the execs figured they'd 1) Cash in on the AI buzz, 2) Save money by cutting artists out of the picture regardless of the loss of quality, and 3) confuse the viewer with bright lights and quick cuts. On a side note, those monkeys are indigenous to Japan, and they spend a lot of time bathing in natural hot springs there even in winter.
What i will never understand, is why is it strictly forbidden to touch up an AI result?.. because that is seemingly the case from the get go. Would it go up against some secret digital progression codex?
The trucks wheels change in every single shot. Sometimes they blend and sometimes there are three, then two and one very large compared to the others. The edges of the trucks waver, as if in a fog. This is really shoddy work.
couple more weirdness (you might have noticed but not mentionned) the two reinders, the first ones that appear, also have their horns that merge together as they move in the duckling shot, you can see the trucks undergoing mitosis (a new truck emerging from one of the ones present on screen) the super close up where the wheels don't turn, you got carlights floating beside the truck. Just floating there, moving as the truck goes forward In japan the two trucks in the back merge into one as they roll And then I gave up because that stuff looks so bad, uuuugh We should just shame and mock any corporation that uses AI like this. Make them laughing stock and make them loose money.
It definitely worked in terms of marketing. You had us staring at the Coke logo for nearly 20 minutes 😬 There’s something about these AI generated videos that our brains kindof pickup on but we’re often not quite sure about on first glance, like the uncanny valley.
heavy lifting with after effects since anything with coca cola was unacceptable. so many layers and masking to hide the mess ups, the two seconds on each shot makes sense that the AI would fall apart after that. There are 4 different AI studios that made different Christmas Ads. This one doesn't have people in it.
One of the best videos on AI art and what is wrong with it. I will be sharing this with others so they can learn the problem with AI images. You have a good eye for detail which many do not so it helps with your point of view. Prompts are a big problem with AI the same prompt can give different results. I think the trucks here are a good example.
Most people aren't going to pay enough attention to see much wrong. Make sure they see it, and tell them about the children of animators who will not be getting Christmas presents because Coca-Cola decided they would do a good enough for who it's for pos commercial.
I regret not making an Arctic monkeys joke.
Remember to check out the description to pick up the Essential Topology Guide for 40% off until the end of November.
Please, explain what that joke is, thank you..! :) I'm a fan of yours and Arctic Monkey, so yeah... LOL ...would love to know, cheers! :D
After decades of research, marketers have finally found a way to make advertisements even more of a waste of your time
I love the 5 star ad between all this chaos
Pepsi has the opportunity to do the funniest thing and remake this for real.
@@FeRReTNS Pepsi. "The real thing".
I think all of them will follow and do AI.
Lots of quick cuts in hopes you don't notice all the crap wrong with each scene.
Bingo.
no, its because the AI that exists to do these videos only can render small segments into motion. I've seen a bunch of clips of how to do these with various software and its basically just taking a still image and animating a couple of seconds of motion into it. These AI clips are run by prompts. You have to be very specific just to get an image that is reasonable. And it can only extrapolate a very small amount from that point on before it gets beyond the prompts and starts to get really wonky. As is they probably had to have the AI animate each of these images a thousand times just to get something this good.
As long as I know the art is AI generated then I know it is bad
but if I can't tell
then it won't be bad or good until I know if it is ai art or human art
Like the guy above said, the tech can't generate anything longer than a couple of seconds. They have to edit a bunch of short clips together.
No. I mean yes. It can a bunch of reasons. The model used might begin to quickly fall apart and introduce to much crap wrong, very quickly. Some of the shots might in fact be 3d scenes rendered and then taken a single frame to let the AI sauce create a believable 30sec moving shot out of it. (serious.)
The actual main reason for it? Trying to make a advertisement with this tools is like trying to paint a painting with a mouse cursor. And no, I do not meant in Paint. But trying to make the screen cursor go out of the screen and pick up the pencil and paint on the canvas in the room.
Generating still and moving pictures with AI? In the now 'traditional' meaning of it? You type text into a prompt. Like this very comment is done. But this comment is not trying to recreate a cinema grade advertisement for cola. What would a machine-learning prompt look like? Have no real clue how a cinema grade advertisement for cola would completely look like. But can give a very short and badly done one: Red truck driving on a winter road. Time of day night, snowy, calm, sooding, $$$$$, imput emoj here.
Try and imagine crafting a text prompt that can be used! You have to write up a strange machine-learning oriented scrip for every scene. Shape and coerced the input prompt until you get something usable. And repeat this endless cycle of getting the machine-learning to produce a scene that is what your after. This is not magic. Anyone having done 3d modeling or anything like painting/shooting movies/pretty much anything requiring a skill to make something look realistic? We all know that it is all a big amount of fakery. Polygons and paint strokes. Even if depicting a real event and real objects? It is only that. A depiction by means of something else. Machine-learning being marketed as AI is simply that.
The amount of control some machine-learning creative code can do is simply limited to what it has learned. And the control we have over it is rather dull. If a pen or paint brush is not working as you want it? Change it. If a 3d program is not behaving as you want it? Then with source code change it. If a AI Machine whatever is not behaving as you want it? Then good luck getting it to behave. With almost no re usability. You have to regenerate everything for every scene and change. And I mean not re-render the lighting. But every pixel and so shape of the hole world is going to change. This AI machine whatever codes can not even be relied to make trucks have a look coherent to each scene. You can probably try and tell it to make: American, Truck, 18wheeler. And yet you are not really in charge here. And instead get some European truck. You can try and describe and input as much as you want. It simply is not going to do it reliably. It is so out of the control of the 'artist' as you can get it. Simply to much is up to randomness in the code. Machine-learned-code. Gibberish. We need actual AI to understand 'AI' code.
Meanwhile a photograph or voice recording from 1700s can be put into a scene or used as is, anywhere. Edited etc, etc. Converted to digital or analog back and forwards. A machine-learning prompt might be good to try on a different model? Plain readable text. Not actually something worth alone to be reused. Not even sure if you can make a machine learning model recreate the same output twice. At least when it comes to this scale of machine learning. Quite allot of things need to be made up on the spot to create something out of tin air. And it is out of tin air. It is using what is learned. But how do you pick what to use of what you have learned? Creativity. Intuition. And that is exactly what AI lacks. At least while also keeping track of everything.
One can not go into a AI-generated scene and change something. Since if you try? You are going to regenerate everything and something completely different is going to exist the next time you hit render. At best you can feed the AI with AI created things? And hope that it dose not learn from AI mistakes. AI is pretty much perfect for Photoshop and such effects. We have had AI for a long time. We only called it Photoshop. Trying to use Photoshop to make a Cola advertisement alone? You would pretty much make a animated moving image. As is trying to use 'AI' to make a moving image. You get the results based on that cause. Cause and effect. Text prompt and weighed values? That is what the 'creative' person get to actually control. Then it is up to the AI to make something out of that. Try and use Photoshop with a text prompt. That is what this is. With no ability to go backwards in the processes. You get a single text prompt and Photoshop makes a hole scene out of that. You have literally no control. Only hopes of something good. You might as well outsource the work to someone else at that point. Being a artist.
How many, and how much time was used to generate this scenes? It is very likely a embarrassing number... And if not for the panning camera? The scenes are rather static. Moving heads. Moving trucks. Very short shots. All because it can not be controlled.
You could literately write a machine-learning prompt and send it to someone that do this kinds of projects for a living 'the old way'. And you would actually get a better output. Since I do not know, maybe using Photoshop alone is a really bad idea when making a short movie/advertisement meant to look real or good in any way? Use the lasso tool to make a hole movie!
Maybe I'm wrong. Maybe this AI Cola add was from begging to end generated. Without editing of any kind. And the AI also had the pleasure to export the finished file and upload it to... wherever ads go. Or this are a bunch of random scenes stitched together. Hardly possible to make a single 30sec scene look good enough to fool the general public to not see it as a bubble.
If I was to draw or model/animate a reindeer? Or really anything? I would look up and find images and such to make it look right. I wounder what data machine-learning uses.
It can not be looooots of data to do what a artist would be able to do as well without image references. Almost like humans are better then a machine at learning. And without the need of braking laws.
If they wanted to save money, reusing an ad from 20 years ago probably would have been better and less hated than this.
As someone who works in an arthouse this has the fingerprints of junior artist's first after effects job all over it. All the branding are just PNGs tracked in, hell, all the coke bottles are just PNGs. All the lights turning on are simple masked altered footage and you can spot the comped shots clear as day; they're egregious.
Coca Cola cheaped out on this, big time.
But think about all the artists they didn't have to pay! Huge W for the CEOs!! Might save enough to buy another Bugatti.
There are actually macaque in Japan, that chill in Hot Springs. So that is the only part of the Ad that makes sense and looks okay.
Yeah, I was going to point out they are supposed to be Japanese snow monkeys that sit in hot springs in The North of Japan.
except Japanese macaques have different faces... still a crap part
It's incredibly apparent that nothing moves in three dimensional space, but rather in the style of those animations that convert hand drawn pieces of art into motion.
Ai animations are suffering from being blurry and also objects with ai animations never actually move, they almost always seem to morph around the screen
"Cars can only go two miles an hour so my horse is sure to win"
It's over, bro. The 3D industry is finished and people are in varying stages of denial over it. The last of which is criticizing the "art" of commercials for corporate clients.
@@wrong_planet wdym it's finished? How else would vfx be created?
@@jeanclaudethedarklord6205 When's the last time you paid a team of people to do in six months what you can get in ten seconds for free? The 3D industry is done. Commercials, movies, games, television shows, art commissions - already being outsourced to AI after being outsourced overseas. People do not notice a difference and they do not care.
Now I wonder when we’ll get AI-generated reaction videos with all the comments generated by bots.
Finding a real person on UA-cam will be like finding a needle in a haystack.
It's already in effect. AI generated channels with AI / bots generated comments, and artificially inflated / bought for subs.
Here on youtube it's already all over the place.
Love that the hand has 6 fingers. Jesus christ 😂
I too, as a global brand, want 45 different shitty art styles in my advertising.
I wonder if the AI company paid coca cola, and not the other way around.
Wait till next year 😊 ... even more companies will create their own commercials in-house soon 🎉
Also, its an advertising stunt. Everyone is talking about it and some people think no publicity is bad publicity.
This signals that quality is not important anymore. This is exactly why students in tech and arts are dropping out, who will hire them?
There are actually monkeys in Japan that are taking hot baths when it is snowing in higher altitudes
Yeah I've seen them. I just didn't think this was supposed to be hot water. Not enough steam.
I'm still not sure what monkeys have to do with Christmas
Agree, it looks like cold water but they seem to sill have fun😊 also japan is definitely not famous for celebrating Christmas. I think they just did all that with AI for getting attention and it works
@@DECODEDVFX They do look iced in. Maybe someone should go help chip them out. And why are the hot springs touching the asphalt/tarmac?
I find that using AI in advertising like this accomplishes the opposite goal because the more times it’s shoved in your face, the more you notice the weirdness and it becomes more unappealing. AI stuff works best when you only see it once in passing not over and over and over like Modern advertising
Let's hope the trend continues and everyone eventually stops buying this shit.
Is this a prank? This advert looks incredibly janky and hilarious. From an informed eye there's so much wrong with this xD
Exactly what I thought.
As a VFX guy myself, a big company like Coca-Cola should be very careful doing an AI commercial like this, because AI sources the imagery from multiple sources. The risk here is some source elements getting mixed into the advert, that has come from an artist or company, that Coca-Cola haven't had permission to use. Could result in them being sued very heavily. AI imagery and video still has a lot of question marks over the head, and with this new tool soon coming out, that supposedly will be able to unscramble the AI and source the elements. When that happens things are going to change massively.
Well, the actual risk is infinitesimal; I've made an experiment with MidJourney, prompting it to recreate some famous paintings. With Mona Lisa it was true to every detail, even in the background. Picasso's Guernica was recognisable, but various details were missing and patched up inconsistently. Munch's Scream had the overall composition, but the screaming character was created completly anew, more realistic and not in the Munch's style at all. As you can imagine, it went downhill from there. It couldn't recreate even the basic composition of Rembrandt's Night Watch, it returned a generic period scene in the "style" of Rembrandt. So unless something is as iconic and overrepresented in the sample base as Mona Lisa, the AI is not likely to recreate by accident any details of any single piece of art sampled in there.
The polar bear is included because it's a throwback to the most successful Coke marketing campaign ever. No other reason.
Some of the animal shots apparently John Carpenter originally deemed too cute for the theatrical cut of The Thing, and they eventually ended up here. An other testament to the power of film, scanned in beautiful 4K about 40 years later.
I don’t like AI ads, but it is crazy how far AI has come with like stablediffusion. Just think how far it’s come in just like 2 years, from the will smith eating spaghetti ai video to this.
This is a totally different AI coke ad than I've seen people mocking on social media, the one I've seen has coke bottles changing size and shape in people's hands.
@5:39 - You can see a tracking error with the logo on the second truck - definitely looks added in.
@8:13 - The back of the second to last truck has a messed up logo. Probably one they forgot to replace with a "clean" one.
Probably. I churched out about 30 different AI thumbnails for this video, and it messed up the coke logo almost every time. And that was using Flux, which is one of the best algorithms that's particularly praised by AI bros for its skill with text. 🤷♂️
The monkeys are Japanese macaques, the only monkeys native to the archipelago, who are know during winter to enjoy spending much of their time in hot springs. The ad may have a lot of problems, many of which you pointed out, but showing the macaques in the hot springs just after the trucks passing trough a Japanese city is a very intentional human choice derived from reality, and not one driven by AI madness. Also, side note, japan loves Christmas and they even have their own unique traditions built on top of the American/European christian ones.
I know that KFC on Christmas is a bit of a tradition there.
Who doesn't like Happy Fat Gift-giving Man?
They started adding christmas lights all over the country like a month ago, it's a very different christmas, buy they do kind of celebrate :P
I'm pretty sure those ARE supposed to be mice. I.e., everyone gets along at Christmas.
It always makes me perplexed. These ads used to be so pixelf***ed for accuracy and perfection. But as soon as AI bros got in there.. it's like all kinds of quality control went out of the window. So many of these issues could probably have been corrected by simple compositing and paint overs or just regeneration. But the only postwork here seems to have been the logos...
I love seeing Santa criticising Coca Cola's christmas ad😂❤🎅
How was this level of quality allowed
You are being so nice with this shots
You say "This bunny, I supose it's ok", but it's fur is completely fckd up
"This trucks are not bad", but one truck is emerging from another one!
And so and so lol
This is spit in all our faces! How many hours did we all spend because a water drop did not seem realistic to the client and they threatened not to pay us, unless we make that water drop look more real than real. Or calling us out if we could not match a multi million dollar production, for a few hundred bucks they are willing to pay. I wish someone would put Cocaine back into Coke. That would make one hell of a commercial!
Also. Why do I get the feeling that what was fueling this ad... Was another sort of coke.
I bet the AI didn't do any of the lights "turning on". They just "turned them off" in post by dimming and desaturating them.
@@dack42 that's my theory too. I feel like this video took an awful lot of manual work to meet the brief they were given.
The hand filling the truck with Coke had five fingers!
Six.
"I don't know what the hell monkeys have to do, with Christmas" - DECODED
The animals' skin and muscles don't move! They just morph from one pose to the next! It's like I am looking at my in-laws!
6 fingers. It's said the demon leaves his mark in everything he owns
You can bet after this was unveiled to the board someone will be getting a P45 in their company christmas card
I'm pretty sure all the lights were burning bright in each generated shot, and they manually dimmed them down at the beginning to create the "turning on" effect.
Ok you talked about it 20 min. So comercial succeeded pretty good.
I just hate how big companies just shuts their eyes to AI's flaws but if it was made by some videoprapher or CGI artist they would be pin pointing the tiniest thing to make it perfect. And let's agree it always looks unsharpened, blurry and low resolution.
I hate it too but that makes complete sense.
If a single artist remade this entire ad, then first of all it would take 200 times longer to make. And since humans are supposed to be better than AI (that's what most artists seem to think), then companies will expect little to no mistakes.
An ad like this, only made by humans would require a lot of resources to make perfect and respect deadlines.
AI can do that for essentially free, and the result is decent, mistakes are tolerated.
Give it a few years, it will remain cheap but the result will be perfect. It will still require human input for artistic direction, but much less than required before.
AI video is just a few years old. The first generation is never the best.
I get what you're saying, but I think they should wait this few more years before it's perfect, then. And this kind of work are rarely done by a single artists but done by a hole studio.
Those are HUGE companies with a lot of money. Hiring a studio would not be a problem and would result in a better result.
But we need to understand the context here. This is a partnership of Coca-Cola with Dall-e. So it's nothing but business here. The money they're making with this partnership may be worth the poor quality.
no way this is a real commercial
I laughed my ass off at the mutant animals. The truth is 99% of the public will not notice a thing because grandmas will see a bizarre ai image of some kid in africa building a space ship and they will think it's legit, AND they get free marketing from the backlash for using AI. I think it worked pretty well.
It's a Centiduckling.
"They went for the AI thing because they know it will get headlines."
Well... The effect on me and my friends was pretty great. Now I refuse to buy Coca Cola and I showed it to my friends that normally don't even care about Coca Cola Christmas ads and now they also boycott Coca Cola.
You mention how the logo is always crisp and clean - but if you look at the shot with the driver, the logo is out of focus in the background and is, really, the only time it looks odd, especially the O on the word 'Coca' which is clearly independent, but in the out of focus shot, is merging with the c next to it!
The ducklings keep losing a leg.
AI is just a relatively new-tech tool which is going to get better and better in the future. While I like the MESSAGE of Coca-Cola about united people as a family in sort of magic and prosperous days, kindness and stuff, I dislike the high level of imperfections in todays AI. But I'm aware that could and will be fixed gradually, so it really does not bother me that much
I don't know if the first bear in the tree is meant to be a polar bear. The head shape is wrong, and while it may not be intentional (because AI), it looked tome more like a Kermode bear, which is a white/off white subspecies of black bear.
This reminds me of the amount of work required to make morph and warp effects convincing at post houses in the late 90s. I'm guessing - beyond all the work trying to prompt this amount of consistent-ish footage out of an AI - the ad agency artists and producer took ages checking, tweaking and trying to stitch together this ad from the mountain of generated, mutating footage..
The deers' necks get longer and longer. The hand holding the bottle has 6 fingers. All that is creepy.
There are monkeys in Japan that actually do use hot springs.
Also It must have taken many interactions to get this sloppy result 😂 they could have done it in CG in the same amount of time using a decent asset library
I believe there was actually a lot of post on this. I believe all logos were tracked in, bottles were pretty obvious.
I think some of the shots were comped together as well like the ground popping in when he puts the coke case down its actually just a crappy roto mask that wasnt feathered to get the contact shadow in.
This must be a test to see what they can get away with. There's a Coke advertising exec buying a new house with the money they saved on this ad.....
Welcome to XXI century:
Is it good? No
Is it good enough and cheap? Yes
it is not good enough, tbh
@@error.418 it is not good enough for you and me, but it's good enough for millions of people with... "poor pattern recognition skills" (I'm trying to be very polite here).
@@pawe9082 yeah, and fast food is "good enough" for millions of people, but I feel like we should work to do better for our fellow humans. if we keep letting garbage be the most accessible thing pumped into the system, we just end up with more garbage, and that affects all of us.
@@error.418I completely agree. Unfortunately the big tech still has an aura of cool young dudes with noble mission.
They should be treated like any other industry. I've never seen, for example, Exon Mobile CEO giving lessons to enthusiastic crowds about moral values, bright future for entire planet and similar nonsense.
@@pawe9082 hey, if they're scamming venture capitalists, go for it, just stop leaking the garbage 🥲
I auspect the de-saturated lights "turning on" is not made by the AI. It probably created the whole shot with all lights on, then artists went in and manually de-saturated them in order to create the "turning on" effect.
There actually are monkeys living in similar environments, but not in ACTUAL ice water! :D
Lol at Newtons Cradle of Ducklings. In first deer shot the antlers end up merging as well.
People feel the difference, even if the can't 'see' it . Something feels off but they don't know why.
I think Coca-Cola's biggest problem is that It believes the general audience is dumb enough not to notice all of those weird things.
In the grand canyon sunrise the shadows point about 45 degrees to the left of the sunrise. So, a second far brighter sun behind the camera? That sounds ominous.
the scariest side of the coke ai ads... it worked... everyone is talking about it. The vallée dérangeante (don't know the word in english) is still there and that's probably what they were looking for, that's a really disturbing idea to consider a world where robots are better seller than human. probably something who can generate a "small alignement problem".
It doesnt increase sells
"uncanny valley"
The idea that bad publicity is still good publicity isn't true.
8:10 you dont catch that the front truck has 2 trailers but no wheels underneath at the join. I dont know if trucks can have doubled up trailers like that, but I would imagine they would have some connection to the ground so the truck doesnt jackknife on every corner. also the back trailer has like 6 wheels as if it was a hotwheels toy :'D
OK, well I'm not a vfx artist (maybe one day, hence my sub) but to me, that looked like a big, shiny late-capitalism convoy through uncanny valley. Something unsettling about it, but I didn't really figure out what until you picked it apart.
- Sent from my iPhone
But if we make Christmas commercials with actual effort, how will we have time for all of the DraftKings commercials?
The high contrast… the worst part for me was the truck tires
I know we're all supposed to hate on this, but honestly I'm surprised it says good as it is. It has a consistent visual style throughout, the panning and tracking look like typical camera movements in video, and the continuity errors don't really jump out.
I am curious how they trained it though, like what if someone had to essentially build the whole thing with a mix of real footage and cg, essentially doing the whole commercial traditionally, and then they had to recreate all that work with AI, using something like a style transfer to get the desired Norman Rockwell look, just to say that they used ai?
I'm sure people initially are thinking 'why would they want to say they're using ai? Everyone hates it'... but I can imagine a possibility were there marketing team convinces them they should do something controversial on purpose, because it will get people thinking about coke, and realistically those people aren't going to refuse to drink it just because of General AI bashing. How many of us are currently thinking about or talking about pepsi? Somehow Coke has gotten me to see this commercial twice and I don't have a TV and I have ad blockers up the wazoo.
Edit - oh, I should have waited until the end of the video, he already said it
Ducks walked sideways, perspectives were jank on most of the shots, and tires spinning when they should stop, trucks size not matching environments
The monkeys are totally relevant, the US is a little strange because we have the story with the reindeer, but in most countries outside of the US it's 12 Monkeys that guide Santa's sleigh. And instead of one being an outcast because of his red nose, it's the one baboon who snuck into the group who is shunned because of his red ass.
its cost 5% and makes sense 30%....for some thats a win
The reindeer at the start, had parts of its antlers just appear to detach and float in the air before disappearing (thought it was a tree at first but pretty sure it isn’t)
The shot of the trucks after the monkeys, the front second truck seems to be merged into the rear of the font one. Also it seems to have lost all of its rear wheels
Missed in the back ground while the poor ducks' legs helicopter beneath them: the 2nd truck undergoes some kind of mitosis to become two separate trucks.
With nothing said of the actual commercial - I just hope that whoever thought of this gets replaced by AI
3:19 - I think it's a quite clear separation between the legs. Looking at the white mane thing, it overlaps over the back leg.
That's not a coke ad, its an ad for the AI company.
It’s not a very good one
@kaiser9321 yeah, I agree
I think it's okay to use ai for pre-vis or to find the "look" but for the final product, that's asking a bit much.
Yes. Absolutely. I've used a bunch of AI lately for really early pre-viz. But it should never be in the final shot as it stands.
When one of the biggest companies in the world pulls this crap on the biggest marketing campaign of the year, it is nothing more than a huge middle finger to production houses and artists. To Hell with Cocky-Cola. We should just wiggle our bare arses at them and buy someone else's product. If they don't lose sales over this, everyone will follow suit, and it will be like bullet-time all over again. Thank you for doing this, because I wasn't going to watch their official video and give them the satisfaction of another view.
4:31 can we not talk about Santa 6 fingers.
In the last scene, the snow in the snow globe isn't moving.
@@BMF6889 I noticed another dozen problems with this ad while I was editing the video.
Get used to it, this is what all "entertainment" will be in six months.
It does snow in Arizona. In Japan there are monkeys that spend a lot of time in hot springs.
5:52 the second and third truck are melted together, and then they separate
Yeah, tons of messy stuff in there, and yet they let it pass. This stuff would never fly in a TV show or film. I think the execs figured they'd 1) Cash in on the AI buzz, 2) Save money by cutting artists out of the picture regardless of the loss of quality, and 3) confuse the viewer with bright lights and quick cuts.
On a side note, those monkeys are indigenous to Japan, and they spend a lot of time bathing in natural hot springs there even in winter.
What i will never understand, is why is it strictly forbidden to touch up an AI result?.. because that is seemingly the case from the get go. Would it go up against some secret digital progression codex?
The trucks wheels change in every single shot. Sometimes they blend and sometimes there are three, then two and one very large compared to the others. The edges of the trucks waver, as if in a fog. This is really shoddy work.
8:25 even the cocacola logo got all messed up on the second truck :)
couple more weirdness (you might have noticed but not mentionned)
the two reinders, the first ones that appear, also have their horns that merge together as they move
in the duckling shot, you can see the trucks undergoing mitosis (a new truck emerging from one of the ones present on screen)
the super close up where the wheels don't turn, you got carlights floating beside the truck. Just floating there, moving as the truck goes forward
In japan the two trucks in the back merge into one as they roll
And then I gave up because that stuff looks so bad, uuuugh
We should just shame and mock any corporation that uses AI like this. Make them laughing stock and make them loose money.
3:25 - The left one has a telescopic neck :D not a bug, but an evolution.
They just look like animals sculpted from cylinders
AI produces 4 nostrilled dog - "I don't think most people notice ..." says dejected artist who takes immeasureable pride in his craft - lol
Yikes! I saw the ad and did not pay too much attention, at first. Seemed OK. But now? It should be called _The Nightmare During Christmas._
So I guess animals are the target audience for this Coke commercial?
8:30 the Coca-Cola logo at the back of the 2nd last truck is ai generated
It definitely worked in terms of marketing. You had us staring at the Coke logo for nearly 20 minutes 😬 There’s something about these AI generated videos that our brains kindof pickup on but we’re often not quite sure about on first glance, like the uncanny valley.
except now I feel nauseous when seeing the logo
it's sad their go to studio probably pitched this.
heavy lifting with after effects since anything with coca cola was unacceptable. so many layers and masking to hide the mess ups, the two seconds on each shot makes sense that the AI would fall apart after that. There are 4 different AI studios that made different Christmas Ads. This one doesn't have people in it.
One of the best videos on AI art and what is wrong with it. I will be sharing this with others so they can learn the problem with AI images. You have a good eye for detail which many do not so it helps with your point of view. Prompts are a big problem with AI the same prompt can give different results. I think the trucks here are a good example.
The bottles seem to only be 2D images that just face towards the camera.
This gives me some "lazy intern" vibes.
Most people aren't going to pay enough attention to see much wrong. Make sure they see it, and tell them about the children of animators who will not be getting Christmas presents because Coca-Cola decided they would do a good enough for who it's for pos commercial.