How to Detect AI Images (and why it doesn't matter)
Вставка
- Опубліковано 7 вер 2024
- It is a very weird moment and it is really important that a lot of people have both the kinds of skills that give them good signals for when to get suspicious AND strong abilities to figure out whether things are real or not.
Both of those things, IMO, are getting harder and harder. Platforms (and creators) succeed by giving you content that you /won't/ be suspicious of, and we have more and more tools for fooling the BS detectors that people have developed, requiring new BS detector upgrades constantly!! IT'S A MESS!!!
PROJECT FOR AWESOME MATCHING FUND www.projectfor...
----
Subscribe to our newsletter! eepurl.com/Bgi9b
And join the community at nerdfighteria.com
Help transcribe videos - nerdfighteria.info
Learn more about our project to help Partners in Health radically reduce maternal mortality in Sierra Leone: www.pih.org/ha...
If you're able to donate $2,000 or more to this effort, please join our matching fund: pih.org/hankan...
If you're in Canada, you can donate here: pihcanada.org/...
John's twitter - / johngreen
Hank's twitter - / hankgreen
Hank's tumblr - / edwardspoonhands
“If we vibe with it, we don’t scrutinize it” is the most 2024 way possible to simply explain confirmation bias.
Bubble... What bubble?
(Bubble... What bubble?)
Echo chamber says what.
different lingo, same sentiment
I like the word vibe. It's been very useful... it's got good... vibes.
And it will lead to the downfall of society if we don't fight it
@@youtubecommenter-on9kd Are you saying we need to fight off people's feelings about things, or that we need to fight off AI images, because I definitely want to save humanity but these instructions aren't very clear.
I recently came across a social media post that said, "It's funny how recognizing AI art nowadays is just the same old rules as recognizing the fae in old tales: 'Count the fingers, count the knuckles, count the teeth, check the shadows...' ... and under NO circumstances should you make deals with their kind."
Which is pretty dumb, new AI has no longer any problems with hands or teeth. In a year? What Hank said, forget it, no one will be able to spot anything wrong...
Fairies are AI generated 😮
This is so good lmao
@@KuK137 ai absolutely still struggles with hands and teeth (but it's getting better)
🤔🤯 I like that way of thinking
The saddest part is that the pattern recognition already fails. There have been many examples of regular artists being accused of AI simply because their art style has some of that "softness" that AI generated pictures often have.
Usually because it's those artists' work that gets sampled by the AI
for now, showing our work as we go with progress shots to prove we actually did it seems to be the only way
Particularly Chinese digital artists like Sam Yang because the machines were clearly trained on their work to imitate their style.
Welcome to the post-Turing-test era.
Saw some college students who had to fight to get their grades because their stupid professor asked chat GPT if it wrote their essays and it claimed it did and he believed it
Applying Step 2 to Hank's background has got me thinking he could be AI too. Shelves jutting off at odd angles. Some books are stacked horizontally. And why is there a hammer just suspended there?
Hank is actually a robot, so his own house is therefore AI generated.
thank goodness somebody else noticed. It seemed so obvious to me and yet all these people in the comments are bragging about how good they are at recognizing AI art. Hank is such a troll
i say this with affection
what hammer????
I see a random can of heinz beans
Something really similar happened with a musical artist I like recently. He used some fan-submitted art for his New Year's concert merch and some people called him out and said it was AI-generated. Unfortunately, it looks like he was lied to by the person who submitted the art and he had to apologize. I did think it was an interesting topic for Hank to discuss once I heard about it. The dishonesty in both stories is the bigger issue to deal with. Paying reputable artists in either situation is the solution.
Billy Strings?
@@NaudVanDalen yep
the weird thing to me about the cards is that the post was about the cards themselves not the background. to call them out about their background being AI generated when the post is about the cards not being AI generated is hilarious, it doesn't seem like deal-breaker territory to me... like, who cares if they use AI to generate the background in a photo on a stupid social media post, as long as the card art is made by real people who are getting paid. I don't even play Magic The Gathering, I just love art.
That's happened at the corporate as well. A corporation pays someone to create a piece of art and then that person uses AI for an easy check and now Bungie has AI generated images officially advertising for Destiny 2.
@@better.better Presumably the issue would be that they paid someone to make that image for them. It wasn't a card, sure, but they did pay someone to make it. Probably didn't pay as much as somebody making a card, but honestly who fucking knows how massive companies decide to spend their money? Either way, it was a potential job for an artist that was taken away by someone using AI generation. That's gonna happen more and more over time. And, for good reason, people aren't gonna like it.
Digital artist here. After fourteen years I'm finally pretty happy with what I can make, but it feels bittersweet due to the recent AI advances. How can you compete with something that creates what you do, for free and instantly?
I've seen AI art booths in the wild at conventions and local art festivals. My reference image search results are completely clogged with AI stock photos. Just last night, my partner showed me a new game with an entire catalog of AI generated characters. And these large companies, that truly can afford the real deal, keep getting caught using AI instead of paying artists.
Right now people are still outraged, but I fear that not too far in the future people will move on. Even as someone who looks at art every day, I'm fooled frequently; I can hardly expect others to have the stamina to scrutinize everything they see.
+
AI doesn't create what you create unless what you create is the soulless mass consumer kind of art that isn't really art anyway. As such, "competing" with AI art is actually very simple: Sell Solstice Sun the Artist, rather than Solstice Sun's Art. There's no shortage of artists who are making a killing from monthly patreon subscribers with no guarantee of release schedule or release topics. This is because these artists have cultivated a brand that people value beyond the individual images, whether that's via their personality or via their consistency.
It wasn't that long ago that artists were worried about being replaced by the camera, which was regarded as a soulless machine that used physical and chemical processes in place of human spirit. Now, everyone recognizes that the camera is a tool that expands human creative potential. You use photographs as references; here is a tool that can create hundreds of reference images according to your specifications - whatever subjects, styles, palettes, compositions, moods, or visual concepts that you want to explore - which you can then use as the basis for finished work that exceeds the capabilities of the AI.
The artists who feel most threatened by AI are those who conceive of art as primarily a technical skill, rather than as a creative practice. I think AI challenges artists to produce art that is original, that has something to express, and which is not merely a display of technical mastery. That, to my mind, is a good thing.
@@yurisei6732 Please never speak again
You can't compete commercially, especially as everyday generative AI gets better. But if you create digital art for yourself as an artist... then who cares?
I'm a professional artist who does a lot of contract work, and I'm starting to see "experience using AI systems" popping up in job postings (especially for companies who produce things like f2p mobile games).
The mobile games I think are a particularly interesting case, because the market here is already flooded to the point disinterest and AI images are only accelerating that. Those companies that use AI to generate their gacha waifus are kind of missing the point of the gacha waifu game, which is the idea that this character is "tangible" so to speak, which is something that requires consistency, coherency and intent. People who just want a picture of a naked lady don't need to give money to anyone, they can generate that themselves.
Ah yes, the infamous marvel approach to fix saturation
"what if we made even more shit even faster?"
@@yurisei6732 youd be surprised at how willing people are to throw money at stuff like that, even if its low quality.
Thats why these games exist. They make money. Sure, most people may not buy something like that, but there are enough people out there who kinda just throw money at everything to make stuff like this profitable. Especially when they're producing these games at incredibly cheap prices since they arent having to pay artist which can be expensive.
The cheaper a game is to produce, the less money it actually has to make to be profitable. Kinda an obvious statement, but it kinda matters. These games arent trying to be the next massive hit like genshin impact, they're just trying to get enough people to pay in by using exploitative psychological techniques that they can turn a quick buck.
And these games tend to do that surprisingly well, which is why they keep popping up everywhere. It doesnt take much for them to make enough money to justify their existance.
And you can use an AI, such as Bard to give you the prompt you need for whichever graphic generator you are using. AI telling AI what to do can get you that job! :)
AI art is going to become an artistic skill like any other. It is already integrated into many artistic workflows and pipelines, especially in large projects like games. The one thing missing from the AI art debate is that generative AI is actually mainly used by artists, not random dweebs.
I, for one, would welcome an hour-long hankschannel video about why AI art should maybe pump the brakes. My degree is in ML/AI and even I kinda feel the same way Hank does. It's super cool tech: we nailed the "could". But maybe we need to give the "should" a little more consideration.
Its been talked about elsewhere but as a whole these forms of AI aren't terrible but as consumers we are just mostly exposed to the worst elements of it. For instance this tech is great forreading medical images and helping discover new drugs.
I don't think it has EVER been helpful (in regards to technology) to discuss "should". Because somebody, somewhere, is going to. Doesn't matter how horrifying or scary or immoral it is -- it will happen, because that's how humans work.
So, I think the better question is "how". How can we best prepare for the changes? How can we steer the course of development for the most human good? How can we regulate and prevent dangerous uses of the technology?
We kind of dropped the ball on this when it comes to the internet -- it's certainly not the worst version of it I can imagine, but there's a lot about it we could have made better if we'd gotten ahead of the curve and thought about things like privacy and misinformation before they'd become accepted as normal. And AI is only going to make all of that MUCH WORSE if we aren't careful.
It's certainly been a very interesting journey in being suddenly exposed to a whole side of society that vehemently hates that I use a developed skill that I enjoy for work and that I don't want it to be replaced by a robot with absolutely no benefit to me and all the benefits to the people richer than me who won't have to pay me anymore.
@@darthtace Given the technical capacity, I agree things that CAN happen, WILL happen. But the question for me is more only the legal line of "how do we then make sure that the true creators get paid for their work." Because every bit of training data was made by someone, and computers cannot make things. They can just sample and remix, and when you do that with copyrighted material, you gotta pay the artists that make the machine possible.
@@darthtacecertain tech can be effectively stopped for long periods of time, like development of nuclear weapons. As the requirements for better AI models increase, only so many companies and governments will actually be able to make one that it isnt entirely impossible to slow down or temporarily cease development.
Wacom, the manufacturer of drawing tablets for artists, were also caught using AI generated art in their promotional material recently. And they also stated that they had purchased the art from a third party who had indicated that the art was not AI generated.
I think we are reaching a point where all art that is circulated/sold under the notion that a human made it should be supported by some proof that it was - perhaps a photo of the initial sketch or a timelapse of the drawing process. Everyone doesn't need to see it, but the companies who buy the art certainly need to have it - they can't keep blaming third parties forever, they are responsible for not properly vetting their partners.
Won't happen because "we thought it was real" isn't actually true, or at least it doesn't matter if it's true, it's just weaponised ignorance. None of these companies would ever be saying anything other than "we didn't know" because every possible other answer causes more PR damage. If there was some sort of authentification, these companies would avoid it so as to continue being able to not know.
They're just doing a less insidious version of what the fashion industry does Everytime it's "discovered" that their clothing is manufactured in a hellhole by unpaid children. Corporations intentionally turn a blind eye to their supply chain and then plead ignorance, "it was outsourced to a 3rd party, we'll do better next time." Wash. Rinse. Repeat.
Then they'll just invent an AI that can do timelapses. :/
Yes
@@rin_okami True, but faking a timelapse video is much harder and will give more chances for red flags to pop up. A video has many many images, compared to just one of a piece of art, and the timelapse needs to make sense chronologically. Eventually AI will figure it out, but it will take way longer.
There's an issue in crochet-related Facebook groups with people posting an AI image of "crochet" and linking to their website to "get the pattern". AI doesn't know what crochet looks like, so you'll get lots of posts from people pointing out that it's an AI image. How do some of us know? Because we know what crochet looks like, and the picture doesn't have any in it. There are efforts to explain to people how they can spot the fakes, including learning how to "read" your work so you know what it looks like. But even if every crocheter does this, there's another issue: someone posted a picture of some crochet pattern books they got for Christmas - two of them were clearly AI covers, one of them misspelled "amigurumi", something someone writing a book about it wouldn't do. The problem there is that a non-crocheter was fooled. Now I'm trying to imaging explaining to the entire world what crochet actually looks like.
I don't know what the solution is. I do know that some folks are making a lot of money selling nonsense to the unwary.
As a disabled stock landscape photographer who has been working for years to shore up my portfolios over at Adobe, Getty, Dreamstime and Alamy, I've watched my income stream fizzle within the last year. Mere words can't convey my grief knowing that this income that I was counting on in my Golden Years, will be virtually nonexistent thanks to AI.
Adobe, Getty, etc. used that extremely broad license they have over your photography to train their own models. Ain't that swell?
And they're calling it "ethical" too.
@@fluxophile Oh and also, I decided to close a couple of these accounts but if I do I lose the money I've earned that it is in holding. Dreamstime only pays out every $100 and right now I've got $75 in escrow. They said that if I close my account I forfeit that money! Stock photo companies are the biggest scam going these days. Right up there with mattress sellers.
@@fluxophilepeople should take these companies to court it's al grotesque what they are doing smh I hate AI with every fiber of my being
Stealing voices, art, images, even music, and working on story boarding? Ugh someone needs to stop this it's way out of hand for sure. All ethical artists stand with you and we understand how bad it is.
It shows that there is no such thing as "ethical" training for these systems. Not violating copyright doesn't really mean much when copyright law is so screwed up.
@@keltzar1 yes, it’s not like I was asked beforehand if I would mind if they used my portfolio for AI training. I was told after the fact - “oh, by the way…”
There’s also usually a weird greasy lighting/texture to the realism style of AI art
That’s my biggest flag for AI is the “softness” of the image. Greasy is a good word too.
I describe it as smoothness. There's like,,, no normal imperfections. It's especially visible with fur or hair, but skin or other surfaces are just too flawless
@@sarahp6512 yep, AI images always look unnaturally "smooth", like someone took an airbrush tool to them in Photoshop but overdid it
@@Grey_3438 Yeah, I did a photoshoot with my dad, and I had to tell him not to use AI to touch up the photos afterwards - he was bothered by some graininess in the images, but it just looks so fake after AI. I'd rather have real graininess than fake perfection.
@@whisper_dvm5157 especially in animated AI works. Even if they somehow manage to make it not do weird morphing, it still looks like it's slithering around on screen.
I have an uncomfortable feeling that at some point in the future, we're going to reach a position where no one believes a person is real unless they've actually met them in person.
How long before a new famous person that is seen on screens is revealed to have a completely AI appearance and voice?
That’s a terrifying thought, but I don’t think that’ll ever happen honestly. There’s always going to be ‘some’ giveaway that the thing’s an AI. Especially with video. A person won’t go to the effort to ensure that every single frame and angle is accurate to real life. It’d only take a few errant hairs or a misplaced shadow to give it away
If you're smart, you're already doing this. I have no proof you're not chatGPT and you have no proof I'm not chatGPT.
The Internet is about to become a wasteland.
Buddy, I’ve been there for years now
It would have to be at a point where AI is sentient, or at least is human-like enough to understand the concept of the uncanny valley, and that presents an entirely seperate discussion that is nowhere near the stuff we are discussing about AI now.
@@mjs3188 I don't think being conspiracy theorist-level paranoid makes you as smart as you think it does. ChatGPT has very particular speech patterns and it's pretty easy to tell apart for now.
This is completely unrelated, but I live in Alberta, Canada, and Hank being a Montanan makes him the only online figure I follow who both A) regular experiences the same weather systems as me, and B) puts out same-day content that sometimes mentions the weather.
And it will never stop being wierd to me. Knowing Hank is in the same deep freeze right now really forces you to confront that he's an actual guy who lives in my world, and not just a fictional character from the internet. What an alarming thought.
There are so many problems with generative AI software, from copyright infringement to creatives losing jobs to environmental impacts to people being paid sweatshop wages for reviewing and tagging content for the dataset.
I'm interested to see you do an in-depth look at all of it.
Sadly the problem is still that none of those things end up mattering, generative AI will save companies money and raise profits, so de facto its a "good thing" since money really only driver of what is good and bad in this society.
Only ones that deserve to lose jobs are translators mistranslating work intentionally.
AI is better than 99% of "creatives". Good riddance.
@@vectorhooves7970all you replied with was cope
@@vectorhooves7970 Of course he doesn't he's almost certainly fishing for reactions. Don't feed the trolls.
I think what troubles me most about this AI generation craze, where we’re becoming swamped by endless, mass produced machine imagery and text, is the lack of curiosity it enables. I love art and fiction because I care about the people I share a planet with.
That’s not me bragging or trying to position myself as enlightened, it’s just fundamentally the reason I get engaged in a painting/movie/novel/etc. Because they connect me, however briefly, to the inner lives of their creators. I learn about my fellow human beings: their aches, their obsessions, their dreams, their cultures. The most enthralling works of art can take us into someone else’s world, building empathy by showing us a side of life we may never have experienced otherwise.
But even comparatively mundane, everyday art, right down to elevator music or cartoons on shop window advertisements; the fact that those have some sort of story behind them, came out of someone’s brain, can make them interesting.
With so much AI art, though, it often seems like not even the person who typed in the prompt and pressed ‘generate’ spent much time looking at the image (or else they might have attempted to fix up the bit where the subjects arm fuses with the bicycle handle, or the pedestrian in the background with no face and three boobs). So regardless of how impressive it might be visually or technically, I’m kind of left with no reason to spend much time looking at it beyond ‘oooh, pretty colours’ or ‘hehe, I found a mistake’.
Why should I care, if the person who ‘created’ it didn’t care, and probably won’t be able to identify it in a lineup of similar pictures in a weeks time? And that’s if there even was anyone typing the prompt in the first place. Maybe someone just left it on overnight to spit out whatever, and in the morning it was all automatically posted to DeviantART without the owner of the account even looking at it.
We can make literally infinite amounts of these images (well, at least until all the CO2 emissions catch up with us), but I don’t feel compelled at all to try and look at all of them because I already know what they’ll be like. They’ll be fine. They’ll be pretty. They’ll be generally aesthetically pleasing representations of whatever their prompt was. And they’ll sit on a big pile, filed under ‘oil paintings of anime girls in the rain wearing blue coats standing next to tigers’ or whatever, and no one will come back to look at them because they can always just press a button and generate eleventy bazillion more ‘oil paintings of anime girls in the rain wearing blue coats standing next to tigers’.
It just removes so much humanity and flattens art and creativity into something so impersonal and commodified. It lets us form bubbles around ourselves by rinsing out the influence of other people, other perspectives, other lives. Which is such a valuable tool if you’re in the business of controlling and manipulating the public. And it’s also just boring.
I don’t care if you’re not the best at drawing! Draw anyway! I want to see it! Because only you would draw it that way, and chances are you’ll have a story about it.
i really love your take on this. often, i see sort of faint art elitism in anti-ai arguments, and saying only the deep intellectual meanings in art matter. To me, part on how even the prompter probably won't be able to tell which images they generated brought out what I struggled to put into words. Artist who spent lots of time deciding every detail can tell which is their art and some can even recall how it felt to draw it, and that is a beautiful thing. Just drawing stuff the artist likes has meaning, even if it isn't some cool symbolism or mind blowing commentary.
@@foxclower Thanks for adding onto this! I’m definitely in agreement: the process of making art is valuable in and of itself. Socially, psychologically, culturally, politically it’s important. But it’s also just good because it can be fun. Sometimes art is useless. But the notion that only useful things are worth doing in life is one that ultimately treats human beings like tools, as ‘thing-make-happeners’, means to an end. To be truly humane is to value personhood for its own sake, in all its frivolousness and inefficiency.
We are toolmakers, but we ourselves aren’t tools. We craft machines, but we shouldn’t treat each other as interchangeable gears in a mechanism. We make things, but we are not products. I want people to feel like they don’t need permission to do things that are time consuming, that make sense only to them, that are silly and ultimately serve only to make them happy. We don’t need to optimise that part of us for maximum quarterly output. Just grab a pencil and start scrawling! It feels great!
Brb going to go generate an image of anime girl in rain next to tiger
I really love your take on this and I wholeheartedly agree
@@KD-ou2np don’t let me stop you! go get ‘em, tiger(s in the rain standing next to anime girls)! although while you’re at it, could you also generate 47 more, each almost identical besides minor variations in pose, composition, colour palette and number of limbs? thanks
Honestly, I am of the opinion that we need to do more then just scrutinize images, for the reason you mentioned(the fact it will get better), but that we in fact need to put regulations on the training process itself.
Too late
To me, the real problem is that AI is coming to the fore right at the time in our history where lying is unfettered. And people who accept other's lies while knowing they are unrealiable, or who lie without care about their reputation are everywhere now. Especially online where they expect no repercussions. And all these images are being seen by many who don't have much of a grasp on the real world, since it's outside, where they seldom tread.
I used to know a guy that lied every time he opened his mouth. But he wasn't very bright either, because the lies he told were ridiculous. The really ridiculous part of it was, there were people that actually believed his ridiculous lies. It boggles the mind the stuff that people will believe.
THIS. We have an information crisis. It contributes very much to lived experience, societal/political outcomes etc.
These aren't unrelated things that just happened to occur at the same time. It's easier to pass a lie now because of tools like machine learning which make them harder to scrutinize. In turn makes lying a more rewarding activity, and thus breaks down the social taboo against lying.
There is two outcomes in my mind: you will be able to buy consensus, or the public internet will fall out of favor for information.
Its super scary. Roger Stone has already lied about something he said and claimed afterwards it was AI generated.
The thing that drives me up a wall is that this is killing so many working-class artists' dayjobs. The art that takes your breath away almost never pays its artists well (when they're still alive), but they're able to make it because they have a dayjob making advertisements or whatever that doesn’t completely burn them out, enabling them to build their skills over time. If we replace those dayjob artists with AI, they’ll have to go into whatever underpaid, low-skill, high-hazard environment will take them. (Source: I have several artist friends who have had to start doing this in 2023.) And the people who will be able to afford to develop their artistic skill and share the results with the rest of us will be mostly people who were born rich.
Do we really want a culture where only nepo babies are able to make art? Where the only trained and practiced artistic voices we get to hear are people who don't understand what most people's lives are like? That seems like a bad idea.
that's kinda how it's been throughout most of history, the rich and elite wrote poetry, the poor survived, there were exceptions, but not too many
@@Robstar100 Drastically reducing the number of exceptions seems like a bad idea that could easily result in extreme mass cultural alienation even worse than we're currently seeing
This is what bothers me when people say 'oh it'll only replace the boring art jobs that no one actually wants to do' - art jobs are already had a huge problem of being mostly medium-high experience positions with too few entry level positions. Replacing everything entry level with a robot compounds this problem.
and sure, art jobs don't necessarily pay amazing, but a lot of them, at least the ones I do, pay 2x any retail or other entry job that I could pick up on the side without like, pivoting to learning new skills for an entirely different career, paying for schooling or apprenticeship or whatever.
it's one of those things that's just compounded by the other issues in our society. Like, if I woke up tomorrow with universal healthcare and universal basic income I would not care near as much that my field imploded. I'd still miss the studio working environment because I like my job and would rather I do it than a robot do it, but there are other careers I could enjoy. It's just that I don't want my entire finances destroyed by being forced to switch careers so that some CEO can make more money
I mean, if it's any consolation this is/will be happening in basically every area. I suspect many entry level coding, writing, clerking, marketing jobs etc. are on the way out. Translation as a profession basically won't exist in about 15 years. Etc.
If AI remotely fulfils its "promise" then in 20-30 years, at least in developed economies that are predominantly services/knowledge based, we'll either have something like UBI (so _anyone_ will be able to follow creative pursuits) or we'll face societal collapse.
The technology has amazing potential but _at best_ the short term is gonna hurt I think.
@@amberbydreamsart5467 It isn't just CEOs though. I've seen a lot of popular artists charge lots for simple drawings. Drawings which take time and effort and slowly kill their wrists. You can only do so much so of course if you can't meet demands people will find another way. Everyone benefits from this.
Imagine having technology you can train to do your art, you only need to provide a basic outline.
You could literally preserve your health longer, offer more drawings in turn making more money, and still do art fully by hand for people who prefer that.
Artists are basically letting the CEOs overtake them because they aren't making use of the tools. Imagine how profitable it could be to let a company use your trained AI for lets say 1 month for 60$ or something. Or offer them the ability to train AI off your art for 100$ a drawing.
Artists could benefit from this but aren't capitalizing on it.
They are instead yelling at a brick wall hopeful it will move.
I could be missing some things, but this is what things look like.
So I have this argument against AI art that's not very common - I've looked to see if other people make it and I haven't seen it much - and if you happen across this comment, I'd love to hear your thoughts.
The gist of my argument is this: Artists make creative decisions when they make art. And not just a few decisions, either. The more experience you have making a large creative work, like say, a film, or a novel, or an album, the more you realize thousands of decisions get made in the process of creating it. Most are small, some are massive, but even deciding the first word of a chapter is still a decision.
The difference between humans and AI is that humans can have *intention*. We can make decisions based on what we think will service the greater whole of a work. AI isn't completely random, it makes its choices based on a set of criteria, but it's still not really "intelligent" in the way humans are. If you ask humans why they chose one shot over another for a particular scene, they can give you a *reason*. And AI can't.
So for me it's not so much about whether AI can make good art, or whether it takes the soul out of art, even though those are important things to consider. It's about how, when given the choice, an artist should always choose to make something themselves, because the joy of watching your work come together as each small decision you make builds on itself is erased when AI does it all for you.
AI can approximate the output of art but it can't replicate *intent*, and intent can be the difference maker between something bland and something special, in ways that are invisible to people who haven't tried their hand at a huge creative endeavor themselves.
It's an interesting line of inquisition, but I think too absolute to be applicable. For example, before they've even reached the page, an artist has to make the choice between typing and handwriting, or between buying paints and creating their own paints. Now, you could say that these are more precursors to art than the art itself, but that would also be too absolute -- there are artists for whom the tool is more important than the creation, for example because the method of creation or the resource used is intriguing.
AI is ultimately in the same place when it comes to "artistry" and "intent" - it's a tool that someone may choose to use to speed up the process of creating their art in the case that the tool used isn't important to the art. Someone whose goal is to paint a picture may choose to use mass-produced paints to cut out that work and focus on the aspect of art that they feel is important to them. Similarly, someone whose goal is to make a video game, where the artistry is in the systems design or level design, may choose to use AI to cut out the work of creating textures and sprites, or of figuring out how to program the systems they've designed.
That's where AI is going to be going, an image-generating AI isn't going to replace the artistry of images, it's going to be a tool for people who are making complex forms of art that require images but are not about their images.
i have this exact argument too, or atleast similar. Like sure i can listen to ai generatd music, but for me it wont be as meaningful as listening to music by an older artist who talk about their experiences with love or parenthood or famial love. ai can make music or paintings very well but i just dont think it will have the ability to move me like a human artist would.
Any intent comes from the person using the tool, AI doesn't really change that.
Oh yeah, absolutely
So much of art is meta-analysis
Who made it, when it was made, why was it made, what this person meant, in what period were they living, were they living in war? in peace?
What message did this person intend to portray through the art?
That's why "looks good" doesn't matter to art, it hasn't mattered since someone discovered how to capture light through glass lenses
But with AI all that "more than meets the eye" simply does not exist
All of this fundamental questions about a piece of art are replaced with "an AI made it"
The raison d'etre of AI art is "a tech broh wanted to quantify 'it looks goos' enjoy it"
No matter how pretty it looks, it's hollow behind
AI art is the cumulative result of decades treating "art" and "content" as interchange words
@@roderik1990 The person using the tool lacks the skill required to even know what intent is
I wish we lived in Star Trek. I was a hobby illustrator who was considering taking my skills further and trying for actual respectable illustration work, when repetitive strain injury destroyed my ability to draw. I've been through all kinds of OT and the pain is still too much, so... I'm now coming to terms with the fact that I effectively won't draw again.
AI image generation is so cool. I would love to use it as a tool to stay connected to the world of visual art. I can imagine working together with an AI image generator to create sketches, using my knowledge and limited ability to sketch for short periods to sculpt what I want. I'm also sympathetic to all the small creatives, writers and game designers and the like, who could make use of a helper to visualize what they've imagined, when they may not have the money to loop in a professional illustrator to the project, or who don't have 10+ years to just learn to be a professional artist themselves.
But all these little edge benefits are swallowed up by massive negatives because we live under capitalism. We don't get to freely explore the possibilities of his technology for improving our lives. We're in a life-or-death struggle for money against the corporations, and we're losing. And I cannot help but see the disastrous consequences that this will have for the economics and especially demographics of art-making, and the way that it will make the creative and imaginative world smaller, not bigger.
there are SO many artists begging for work. when a large company uses an AI model which functions by using stolen art, they are not only eliminating potential work, they are stealing from the artists they used to employ
all so they don't have to lose a few bucks to pay a real artist. it's depressing.
Correction: No major image generator uses 'stolen artwork'. That is not how the technology works. You can dislike it, you can recognise the deleterious effects it will have on artists, but if you lie about it you are lending power to AI techbros, who can easily prove to leypeople no theft has taken place.
It makes everyone who is against AI art look stupid when you repeat inaccurate info like that.
@@AvalonisHere art station, deviant art, etc were scraped to make stable diffusion possible. You have to train it on a library, and they did it without asking permission or paying anyone.
@@broodovermind You don't need permission, or to pay someone, for making comparisons between their work and other works.
That's what the training is; the machine looks at the metadata and it looks at the work, and compares it to other works.
Should artists who put things in public places be allowed to prevent anyone from comparing their work to that of another, even if only in their own head?
I certainly don't think so...
used to be true but most if not all of the images that were not paid for have now either been paid for or just using free images that have the CC0 license instead of a pay for it license. The older models still exist but the newer ones have been made correctly to not infringe of human artists work nor using other AI images so not to corrupt the training process.
You also still have models that do use art from artists they did not pay. The ones that did make their own required hundreds of thousands or artists to work for a few months without stopping just drawing images that most likely seemed random to the artists. As they own everything their employees make as it is a "work product" they owned the art made by a human. You only have a few models now that are made entirely from company owned art and not including CC0 licensed art.
The difference is noticeable and the new ones do look better as the artists worked together instead of separately most likely still being paid by Microsoft for doing art on their games instead of art for their txt2img AI.
That is how companies should do it the thing is most find using SD1.5 or SDXL easier as they don't have to pay thousands of artists for a few months to draw images of what to the artist must seem random images. Requiring the artist to give up all rights to everything they drew to the company. So cheaper for the companies thus why only big companies make their own. Big meaning like Microsoft, Meta and Google. Meta due to a law suit that cost them enough that they had to do it this way, Google to get away from said lawsuit happening and Microsoft as they wanted their own. Amazon has their own made in a similar way they just fired all the artists after the job was done instead of keeping them all on.
The problem is smaller businesses will go with SD1.5 and trained models on SD1.5 as SD1.5 used images drawn by humans without their permission though the trained models might or might not have done the same. You still have the issue of if the output looks nothing like the input then does who made the input have a claim to the output? Go to civitai.com to see all the different styles of image generation that can be done with just SD1.5 . You cannot tell me 1 artist can do them all without trying to anyways. Still the base model of SD1.5 used images from human artists without asking permission which supposedly SD3.0 is to fix that issue and remove all those images from the training data set. Just took 5 or 6 years and 8 or 9 stable releases to get to it.
I've broadly seen a major downturn in artist output since the AI art fiasco started. I think a lot of people are feeling discouraged, and seeking other outlets.
We’re having our own Arts & Crafts movement.
This is one of the things I warned about when AI image generators started to hit the mainstream -- that it would have a chilling effect on art.
@@AndiNewtoniangenuinely feels so hopeless rn, i still see art a lot but maybe that’s just my echo chamber online
Really?
I'm an artist and I think for me and many of my peers the issue hasn't so much been that we have stopped, but there are floods of AI stuff that takes focus away from real art. It is always discouraging/annoying seeing an AI account gain heaps of followers for no effort though.
We invented agriculture like 5 minutes ago and now every art director at every company needs to be an expert in AI images and how to detect them so that when their contractors lie to them they can catch it or else they get crucified on social media. It's wild.
“We invented agriculture 5 minutes ago”- World’s first farmer in the year 10,000 BC
@@samlee5549You get the point. Evolution isn't that quick.
@@martijn9568 makes bad point, gets called out on it, and responds 'you get the point'
If the point is 'I have no point' then, yes we get it.
@@DellikkilleDI see that you too, have missed the point. That’s OK.
@@DellikkilleD hyperbole dude
A fairly easy solution if an entity declares it's not going to use AI-generated images (I refuse to call it "art") is to have the submitter include proof of their work. Sketches. Time-lapse. The original file with all the layers. Etc. That will weed out the people pretending to be artists quite quickly.
Hilariously, there are artists being harassed recently, while on streams in the act of creating their art - and people coming into those streams to scream about how it's all AI generated. I don't even care if that's just trolls being trolls, it's insane to think that anyone would waste their time in such way. Or make a bot to waste everyone's time.
I can't help but wonder if some of that is simple ignorance of the art process. Western culture has a habit of both elevating art while denigrating artists. ("Get a real job.")
Why would they do that?
That sounds like something that sounds easy but in practice would create a bottleneck, with some artists resenting having to provide the proof and others inevitably forgetting to document it.
Could an AI be created in the future to mimic this?
i laughed at the bit where you said that we are not going to closely scrutinize each of the images we see everyday, bc that's always been a thing that i do, since however long i look at art. which had helped me navigate my way through a lot of things in my life that i have an interest in; people tracing others artworks, reposting images and claiming it as their own, and now ai generated images. that said, i agree that for the majority of ppl, ppl who haven't spent their lives closely scrutinizing art for the sake of enjoying it, it wouldn't matter really, to try to learn how to spot ai. bc ai will keep getting better, it already has.
So many of us have studied and practiced art - whether digital or traditional. For us it DOES matter. AI "artists" often steal parts of work done by actual humans. And it disturbs me that it may not be scrutinized. Recently, I see so many of them posted on "art groups" where they use words like "digital art" (which AI may be PART OF, but many of us would like them to be labeled clearly as AI). I also see a lot of architectural and interior decorating photos that make NO SENSE. Stairs going down the cliff into NOTHING (I mean, right to where a cliff falls off) to a weird appendage (like a mini-bed) stuck to the side of a larger bed, or candles placed on FLUFFY RUGS or BEDS. IF AI is to be allowed, they need to be clearly labeled as such.
I hope more people care, other than artists. The creative field should have been the LAST thing that AI (and people who used them) should have gone into. But people don't observe, or for that matter, care all that much about quality, so I'm thinking things will get worse and worse as time goes on, as people are discouraged from actually creating their own images, because . . . why?
The first caveman to ever draw a buffalo on a wall is probably seething about how the entirety of humanity stole his idea
"The creative field should have been the LAST thing that AI (and people who used them) should have gone into."
Why are the creative fields better than the rest of us? Why did farmers deserve to get replaced by tractors? Why did (human) computers deserve to get replaced by calculators? Why did woodworkers, blacksmiths and tailors deserve to get replaced by factories?
I have sympathy for artists but I'm *so sick* of their elitism. All the comforts of your modern life are built of the back of other fields being automated but suddenly you care now.
@@Maioubi How often do you use self-checkouts? Or avoid talking to a human when calling a business on the phone? Or do business transaction by app or website?
I hate self-checkouts and while I sometimes do use the phone tree (I hate it - I'd rather a human just answered the phone) I usually end up waiting for a person because I called due to the fact that I had questions.
If you truly want to save jobs, then support the work of humans. (OK, on the phone you're likely to get someone thousands of miles away . . .) Buy from small businesses that might have more of a human touch. Support ethical businesses that care about their workers and practice fair trade, and that preserve jobs of artisans.
I said the LAST thing AI should have replaced were artists because the work of artists should be unique and individual, a culmination of years of experience, practice, and development of a unique style. A machine SHOULDN'T be able to replicate that easily. It's not that those jobs are BETTER (and God knows, most artists don't live in any more comfort than most people - it's like when people assumed during the actor and writer strike that MOST ACTORS were like the ones you see dressed up and on the red carpet, when MANY actors live paycheck to paycheck, some of them having to hold down other jobs) but it's the nature of that kind of work, that shouldn't be able to be done by machines so easily.
@@GoddessPallasAthena Go to your closet (or just look down). How many of those clothes were sewn by a human tailor?
Millions of (mostly) women lost their jobs in the 19th and early 20th century, forcing many to rely on their husbands. So unless your clothes are human made, you're basically telling them all, "Well it sucked to be you but the ease of my modern life is more important."
Think of all the jobs we could have back right now if we got rid of power looms.
Now repeat that for dozens of other fields, including the furniture you're sitting on (poor woodworkers), the cars you drive (poor stableboys and horse breeders), the metal parts of everything in your life (poor blacksmiths).
Talk is cheap. Step up to the plate and start your new ethical life.
Because money. And also capitalism. Capitalists would literally rather spend tens of billions developing a masterpiece of software engineering, made by stealing the work of others, in order to make mediocre image with hallucinations, rather than pay artists a living wage (for a fraction of the cost)
I was not ready for this video to just be over. You had more thoughts and I'm invested, please continue 🥺
hank just made me consider that even outside the moral implications of ai art the "enshittification" of ai is probably gonna be very shitty
Possibly, but probably not. Generative AI are too easy to make and there has always been a strong culture on the internet of breaking people's rules just because the rules are there. That's why companies have always had such a hard time dealing with piracy and homebrewing/jailbreaking/etc. You only need one sufficiently pissed off nerd for anyone to be able to do whatever they like with software that was supposed to be inaccessible.
@@yurisei6732yeah, good look with coming up with everlasting new training data sets unscathed by already AI made content
Not even the AI techbro lord of all techbros Elon musk
Manged to put together a LLM that weren't contaminated to its core by chatgpt gibberish
And that guy has more money than God
To know The future, you need to know the current Naration of AI from the Top Tech Companies.
The narration is always about Replacing their employees with AI
AI Art was unnecessary for human progression, it would be better if it was utilized for medical or other tech stuff,
But no, Making AI Art is their Priority.
What this mean is that, their Priority is to replace the Artists and Designer within their companies,
Replacing people, not create more job opportunities.
Also, Netflix recently caught sending email that said they want to replace Artists in their Animation division,
and they're dead serious about it.
not only Netflix, there are a wide movement to Replace other job with AI as well within the company,
job like, finance, spreadsheet job, analytics, accounting,
Those are jobs that according to a research paper was ripe for being replaced,
even replaced earlier than Artists job, because Artists job is still Highly complex, think Movie or Game.
the End Game of AI is to Replace as many employees as possible.
Therefore they spread lies using UBI as a promise, a promise that they won't work ever again,
UBI is a financial suicide, the whole financial system will crumble,
they use UBI to convince people that it's okay to fire and replace as many people as possible,
Removing the Empathy toward Artists, programmer or musician that get replaced,
and they confince you that by doing nothing you're instantly outclassed those people,
that you're the special one that won't get replaced,
until everything is too late and you're also become the Victim in the end
Ai didn't need prompt to operate, they also already farm a lot of prompt from people,
therefore they can operate automatically.
if the Government was actually struggling to pay Pension,
why would anyone think that they
could afford to pay everyone free money?
did anyone actually Guaranteed that it will happen, that it has a clear roadmap?
No
Therefore, the math isn't adds up,
you can declare UBI as another promise based scam like FTX,
Its not. If quality drops they will look at why. And then start filtering training data. They wont continue to train and make the ai worse. They will stop find the problem. Make a ai tool to filter the img data and then continue.
As a fan of both Magic the Gathering and this community, I will absolutely watch Hank and John sit down and try to figure out a game of commander together.
I really want to know which creators Hank watches on YT. My guess is TCC because they have a lot of shorts? Idek who Hank would be most likely to watch otherwise.
@@jacobsamson2827 I'd be shocked if LRRMTG isn't one of the ones he watches.
That will truly be a higlight of my year
For what it's worth as a hobbyist artist who posts online and wants to sell digital art commissions and is distinctly worried about how much of the market is flooded with AI stuff (and how it's trained on stolen datasets), I would love to hear you talk more in depth about generative AI and how you think it might develop. But I definitely agree that relying on scrutinizing possible AI art isn't going to be the way forward for the reasons you stated and given that I already know of some real artists who have gotten hate for posting art in "AI style" (which probably just means that their work was stolen to train datasets).
I am a digital artist that uses AI. I messaged Hank and here is the gist of what I said.
I need to correct you Hank. Art AI is NOT getting much better, that quickly. Sure the Apps and tools are improving significantly. But by and large, for the most part, the learning is done. Sure there will be better training models made, and as I just said better tools, and more skilled artists. But at least for the next several years what we are seeing now will remain more or less as it is... until the next generation of AI comes along. But remember the time between generations is expanding, as the more we learn about AI the more we realize there is left to learn, and the time between milestones widens. Much as aviation did in its infancy.
Also... the current generation of AI suffers from something called Pink Elephant Syndrome. (named after the saying, you can't help but to think of a pink elephant when being told not to think of a pink elephant). This flaw runs to the very core of what AI is and how it works and can't be fixed. no amount of learning will change Pink Elephant Syndrome. Fixing Pink Elephant Syndrome will require a complete rethink.
If you ask AI to draw for you, for example, an unhatched dragon's egg, EVERY SINGLE TIME it will draw you a dragon coming out of an egg, because it CAN'T unknow something. It is compelled to share all the knowledge it has of the thing that you are asking it to do. No number of negative prompts will override Pink Elephant Syndrome.
Quite simply AI has some serious inescapable blind spots. There are some things it just CAN NOT do.
I use AI for clean up. I get my image to 95% complete and feed it to the AI, turn the creativity meter to zero, so that it isn't stealing other people's work, and ask it to redraw the image just as it sees it. In the act of redrawing a piece of digital art it will remove all the things it feels are mistakes; pixelation, the white halo around the edges, it will balance the values and contrast levels, it will even out the resolutions of all the layers so they match. it saves me hours of clean up, and the clean up phase is mostly just tedious uncreative labour. This is what AI is good for. It is helping artists cut through the tedious parts of their job so they can focus on the creative parts more.
I know many artists that try out AI just for fun, get frustrated quickly, and give up and never go back. It is a trend that seems to be taking its toll. The Art AI websites are reporting a steady downward trajectory of subscribers. Many are reporting significant losses in the hundred of thousands of dollars per day. Many sites are on the brink of bankruptcy and have already froze the ability to reimburse unused portions of subscriptions.
I think all digital artists better get used to AI cause they'll all be using it for clean up soon. But these public AI web sites aren't going to be around much longer. Soon AI will be something companies will have to buy privately to be used in house. It is just too energy intensive. In fact, some small studios may decide it is cheaper to pay grunt entry level artists to do the clean up work than it is to pay the energy bill of an AI server.
I love it when my older relatives share AI images on Facebook just so I can zoom in and find nightmare fuel. There was a huge surge of them popping up around Christmas.
And so many of them are ads for products that don't even exist, too. I wonder how many old people got scammed this holiday season. Probably at an all-time high.
It most certainly does matter, especially to those it directly effects. This is the artists who make content, and those who collect (including play) MTG. The value of MTG isn't the card mechanics itself. The brand was built off of the creativity of human artists in their depictions of the scenes that vastly added upon the game mechanics in how it would look into action. This aids the imagination of the players, which makes the game meaning in context.
A.I. art is fun & entertaining, not going to argue against that. However, some of the trained models was trained off of a lot of intellectual property that entities own. Midjourney, is in the hot seat legally for this, if I remember correctly. Be honest, would you rather read your brother's book that were written by him or a simulation of his style of writing? Give me human artistry every time, to that I can imagine & empathize more strongly vs the output of a complex program. Also, I will always want to know about the creator of the piece of artistic media.
This is extremely well said and I agree
This hasn't been true for at least ten years. WOTC started its artless "fantasy realism" design philosophy long before even Return to Ravnica. Human creativity has been irrelevant to MTG since then, because it's erased in favour of a consistent visual style, the same way that if you're working as an animator for Disney, you're expected to set your own styles aside and make something in line with Disney's modern brand of smooth, rounded, big-eyed, motion-captured stuff. The only place you find art in MTG is in Secret Lairs. Of course most of that is bad, but it is at least somewhat artistic.
"Be honest, would you rather read your brother's book that were written by him or a simulation of his style of writing? Give me human artistry every time, to that I can imagine & empathize more strongly vs the output of a complex program. Also, I will always want to know about the creator of the piece of artistic media."
Honestly? I don't think I actually care. I would all come down to the quality of the work itself. Right now, the quality of AI writing just isn't up to par, but in the future who knows. I think about it the other way: if my current favorite book turned out to be AI written, would I have enjoyed it any less? And my real honest answer is probably not. And I also don't necessarily care to know about the creator behind a specific piece of artwork, I think art can and should speak for itself without me needing to interpret it based on outside material, death of the author and all that. Something truly special will shine through regardless of who created it.
@@yurisei6732what consistent visual style are you referring to?
"Fantasy realism" is just about the broadest bucket you can describe to make your argument work, and I'm not sure what you would expect of the games artstyle given the context of the mechanics and lore.
Just in the last few years we have seen sets designed with art deco styling, neo-tokyo Cyberpunk styling, body horror stylings, and right around the corner are top down traditional horror theme sets, murder mystery themed sets, sets based entirely around anthropomorphic animals, etc.
If anything the creativity behind sets has exploded in the last 10 years, and moved away from the consistent "generic" fantasy artstyle and themes that literally every set had until around Streets of New Capenna's release.
Like the first 15 years of Magic was just "generic fantasy plotline discussing wars fought on various planes"
Thank god for this comment, I’ve been fighting for my life down here. All this support or indifference i see towards AI art really just goes to show how few people are artists or just truly appreciate the medium
Human "artists" submitting AI art as their own work to a client that paid them is probably the 2nd biggest risk with it (1st being artists having their work stolen & recreated by AI). 😢
Personally, I would only EVER be okay with people using AI art in very casual/ armature ways.
Ex: A teens whom likes to draw but can't find a good reference picture could use AI to generate something they can reference for their own work. Or if you want a quick image for a DnD character (or other character you've created yourself and will never monetize) and cannot find anything online that looks right for the character.
HOWEVER all of that bwing said, the AI being trained (created) using art that the artist did not explicitly give permission to be used in that algorithm is NOT okay.
This is so unrealistic but I hope in the future there can be a model that pays creators whose works have been used to train a model residuals. That's the only way I'd be comfortable with any AI images (or any other form of media) being used commercially. And even then only in situations where the alterative would be a stock photo not highly specialized art. It's also such a MASSIVE waste of energy it's staggering
as as I know, adobe has trained it's Ai on stock material. Of course they have an extensive library. There is also some kind of system in place, so this is already being built.
However, lutilda, I am an artist and I respectfully disagree. Artists have been "stealing" from each other since forever. There is even an art form called appropriation.
The problem is not remixing it, the problem is plagiarism and where we draw the line. We have not really or effectively been able to protect an artists intellectual property even from other artists, however what we can do with Ai, is to create a code of good conduct. You have already found a good use for Ai yourself in your example , and I will not go into detail how that could be considered unethical, or ethical, from other angles. Because it doesn't matter, because it will happen anyway. Unfortunately in many ways, and fortunately in some other ways.
Ai is just a tool. And we have the opportunity to make it an ethical one.
One unethical part of the D&D character bit I want to point out is that, in the past people would commission an artist to draw their character for them, but now with AI some of that traffic of potential patrons is just going to the free AI art generators, cutting the potential income of artists. That said, this argument is a bit weak, as it's the same basic argument companies use to justify why potential lost revenue is 'stolen'.
@@futuza Ah, I guess I was thinking of it only in the context of a substitute for taking existing art and photoshopping it to be more like the character (ex: changing the eye, skin, hair color), as I know people (esp children/teens) do often despite it being less than ethical to alter someone else's art. The argument I've been told for that is that it's for "private use" so it's okay. I think permission is needed from the artist before something like that is done, so AI art (trained only on art that was approved by the artist to be included and the artist was paid for this use) could be an alternative to that use case. I didn't think of it as being a substitute for paying for a commission. Good point to call out!
@@AgeOfGoldenSilence Very well said! I didn't consider that part of the conversation. It's a very valid point.
It's not often that is disagree with Hank but this is one. It does matter! This is the age of information, and these tools are powering more and more disinformation.
These days more and more things are privatised including news sources, and that means information which is hard to spoof is becoming more and more important. Images used to be critical for that but now with unmoderated deepfakes and AI generated images it's far harder to find trusted information, not to mention what it's doing to the artistic communities in in both the professional and hobby fields. Already artist are loosing potential customers because of these tools based on their own stolen work.
There are things like Nightshade or other similar algorithms that can be used to poison the dataset and protect your work as an artist. The more artists that do this and the more tools that build this in before publishing the work, the less source data the AI has to work from and the worse the models get. This is impacting people's jobs so while it's not going to go away, we're not going to make it easy. To the average person this doesn't matter, but in a decade or two it will when people stop bothering to learn how to create traditionally and leave it all to the algorithm to do it for them as we lean ever more into a consumer society. Thus, it does matter in the big picture, maybe not for the individual. There is joy in creating things; much like the transition from analog to digital, small but important things are slowly lost over time. Where is your validation as a creator when the power goes out, the internet goes down, or it's no longer affordable to subscribe to whatever AI-gen tool being used?
Remember Net Neutrality and how we stopped paying attention long enough for it to be killed? When we begin to think the big picture doesn't matter, it comes back to bite us when it's too late to fix. We're already at a stage where creators have to cater to algorithms because user preferences have largely been removed in favor of engagement metrics instead. The internet is becoming less interesting by trying to over-cater to people. The world will be a sad place when we only creating only for purpose and not for pleasure. At that stage we start losing what it means to be human. Why bother learning new skills if a tool or software can do it for you? Why question normality or bother to ask questions when an AI starts telling you what to think? Why create new content when you can just react to it for more views instead? This becoming the path people are on concerns me greatly.
Nightshade and similar tools are adversarial attacks that only work temporarily - the more images its used in and the more effective it is, the more quickly it will become useless. They work against _specific_ model infrastructures by introducing noise that only those models will do anything with - but the next models will not only not be affected, they will be harder _to_ effect because they'll have learned how to ignore such changes.
Also, the effect Nightshade and similar tools have gets nuked by compressing the image - image compression algorithms are _specifically designed_ to get rid of details imperceptible to humans, which is the method those tools use.
Hank, if you ever want to come shuffle up and play some games of Magic: The Gathering...
The most important thing to consider about an AI image is not whether it is good, but whether it is good enough. Good enough ends careers.
I'd still suggest whether or not the lawmaker and enforcers consider it legal is a good place to start.
I sincerely hope they don't and I sincerely hope those who stolen from real artists are forced to make recompense. As well as those who've made use of this garbage algorithm and been stupid enough to put their signatures on it 😤
Times change, people change too
@@bleh329 Legality only matters if it's provable and enforcable. Which means it only nudges the bar of "good enough".
"Good enough" killed the careers of nearly all woodworkers, blacksmiths, farmers, elevator operators, phone operators, horse breeders, lamplighters, typists and tailors.
@@bleh329 'stolen' so, is studying another's style 'stealing' now? because if so, literally all art in the last few hundred years is stolen. Its painful how many of you have this backwards idea of ownership.
I've grappled a lot with the topic of AI generated art. I've been doing digital art myself since I first got a graphics tablet a couple of decades ago when they still looked like beige etch-a-sketches, so I know the struggle of building up those skills, carefully studying other artists and trying to work out how to make the pretty go. I've done commissions but mostly I just draw for my own fun and because I want the final image I have in my head to exist in the world to share with others.
So I'm at the point of AI trained on artists' work without their position = bad. AI replacing the paid work of artists who have spent years honing those skill sets = very bad. Those are both clear to me.
But maybe AI creating a forest for me to use as the background of an image when the part that's really important to me is the figure in the foreground isn't so bad? Maybe using AI to generate an image of a goblin with a purple hat for my home D&D game when I otherwise would have just googled for something and showed my players the result without any money changing hands isn't so bad? Allowing one of my friends who hasn't ever been able to draw to describe her character to an AI and tweak and change the output until she gets to the point where she can excitedly say, "that's my half-elf ranger, here's exactly how she looks in my head!" honestly feels pretty exciting and empowering.
I want to live in a world where we can use AI tools for cool stuff and be excited about the possibilities they represent, but the state we're at now and the way corporations seem to want to continue is so ethically grimy that it feels impossible.
I look at it the same way I look at Canva. Canva is a thing that makes graphic design easy. Like, stupid easy. (They even have an AI generator--which is a can of worms I won't open cuz idk what they've trained it on) So--BOOM--graphic design is accessible to anyone who can use a drag and drop interface. Which means we don't have to pay those pesky graphic designers anymore!
So is that bad or good? It's...kind of both? Because it both devalues professional graphic designers while empowering people who just want to make a flyer for their bake sale. It's good because it technically gives graphic designers another tool to optimize their workflow, should they choose to use it. It's bad because now that everyone can work faster, clients will want to pay them less or not at all. But it's fine because all the design assets were made by human designers who've been paid (I assume). Probably not very much, probs because they live in a country with a low cost of living and will continue to be paid far less than they're worth, but at least someone was paid and it was all done by the book. It's unfortunate all around, but that's capitalism. Or something.
This is a whole new beast. Now there's thieving and and the devaluing of all creative pursuits in general, it seems. At least before somebody was being underpaid. Now nobody's being paid at all except the people doing the thieving. Unlike with Canva, using generative AI to make a profit is supremely unethical. Though there's nothing bad about using AI generators for honest fun. But it does mean that people really have no need to ever, ever pay an artist to draw their D&D characters for them again, outside of a strong personal desire to support an artist, because they can get the product for free.
Digital art is constantly growing to include more tools to make things easier (including AI, long before all this) and it's always been my stance that it's not cheating as long as you're honest and not stealing. I used Canva to make my mom a flyer in an hour because I couldn't be bothered to do one from scratch. But using an image generator trained on stolen artwork puts a bad taste in my mouth.
idk it's like eating at Chik Fil A. You're obviously not a bad person if you do, but I'd still rather not give them my money.
Absolutely agree, Joanna. There's a lot of nuance that seems to be lost in this discussion. My sister-in-law can't use her hands precisely due to disability and it's really empowering for her to speak to make worlds and characters that were stuck in her head unable to draw.
@@lbon5479it’s not stolen and never will be. Stolen means loss. Nobody lost anything
It's bad for the artists losing work to AI, but I feel like it's not “bad” for society as a whole. It's not the duty of society to make sure artists stay employed, it's the duty of artists to provide services that society wants to pay for, right? If society prefers cheap and accessible AI art over human art, then tough luck for artists…
So you weren’t interested in an artist doing things for you, but an AI that is free you are?
At any point your friend could have commissioned and talked with an artists in the exact same way.
Is using AI somehow making it feel like you created the art?
it’s equivalent to you asking an artist to paint you a picture of a house and telling everyone you painted it because you told the artist to create it.
Hank, I would love you to talk more about AI images and specifically the negative effects of their use and also about how they train their AI on reference images that artists never consented to be used for that purpose.
I really believe that this is a current problem that is causing real active harm to the artists of the world. Some people say AI isn't good enough to replicate real art now. But even if that is true, it is still currently, today, causing harm to artists who make a living off of their artwork or who do it through passion.
Thank you
(edited for grammer, it was pretty bad lol)
+
Historically our society has never required an artist‘s consent to use an artist’s work in almost any way that we like.
Generative AI doesn't cause any problem for artists that the existence of artists in low cost of living countries doesn't already cause. If you're an artist in a place like the US or the UK, and you've been making a living off your art, then congratulations you're someone who'll still be able to do that after AI becomes the norm because there's a reason people are paying you for your art instead of buying functionally identical art at a fraction of the price from a Vietnamese artist who can afford to charge far less.
This matters exactly to your point in that we should always be scrutinizing when we can. This is a slippery slope of ethical generative AI and when we vibe with using things okay, we slow down the rampant use of AI without guardrails. Without guardrails, artists tend to lose out., truth tends to lose out. I don't care about MTG using AI as a concept, I care about what the AI is pulling from and what is the truth behind the generation.
I'd be interested in a longer video that explores the issues with generative AI. Also, your hair looks amazing.
This isn't the first time even recently that Wizards of the Coast have gotten into trouble using AI art. A recent D&D book included AI art in it and they were called out on it. They ended up getting real art to replace the AI ones, however, those with a physical book still have the original AI art.
If you want a shorthand for "why it's probably not a good thing" when you don't want to go further into any detail, just saying the tech is almost entirely predicated on theft is a pretty succinct way of describing the issue at hand and why most artists are not fans of how it's being used or how it was created in the first place. I encourage you to use "image generation" instead of "art" though, because that shit isn't art and never will be. I like many others am interested in your perspective as a non-illustrative creative and general science/technology literate content creator if that's a video you decide to make in the future. Thanks for continuing to voice your concerns and bring awareness to the broader public.
The issue with AI art right now too is that it's plagiarizing a lot of photos and other people's artwork it finds on the internet; it's not strictly using stock photos (and paying for them) or public/free to use illustrations.
Edit: Also yes, it's frickin' freezing in Montana right now.
I am looking forward to Hanks take on generative AI, because there are *so* many bad takes on it
I like what he said in this video, how image generation tools are super fun and interesting , but we're already seeing some serious issues with how they're being used. Hank is great at being balanced and insightful at the same time like that.
This makes me nervous! Will my takes be bad???
@@vlogbrothersMaybe, but in a delightfully Hank Green way.
Hank Green, known for having only the hottest of takes
@@vlogbrothers If your take is "LOL new thing good/bad," then that is a bad take. If your take is even slightly more nuanced than that, then you are already better than like 90% of the internet.
The Wacom story that’s similar to this is more wild because AI art is counterintuitive to the product they sell (drawing tablets)
Hank, can you do a deep dive into the negatives of AI image (and other media) generation? I'd love to see a critical take on it specifically from you.
When Hank mentioned the really rapid temperature drop I felt that... It isn't as bad in MN as it is in Montana but the high today is in the teens when for the past two or three weeks it's consistently been in the 30s. I made the mistake of not dressing properly to go and get the mail yesterday... not doing that again.
I'd seriously appreciate if you'd make a video going a little bit more in-depth into why so many artists are very wary of these text-to-image models as they're currently being used - people are of course very excited about new tech, and there aren't many people outside the world of art talking about its issues.
It is an interesting topic. I know many artists that try out AI just for fun, get frustrated quickly, and give up and never go back. It is a trend that seems to be taking its toll. The Art AI websites are reporting a steady downward trajectory of subscribers. Many are reporting significant losses in the hundred of thousands of dollars per day. Many sites are on the brink of bankruptcy and have already froze the ability to reimburse unused portions of subscriptions.
I was one of those skeptical artists who didn't find many redeeming qualities in AI, and I still stand by SOME of those initial reactions. AI isn't capable of the effortless magic people seem to attribute to it. However with the right skills and experience you can use AI to do the same work you were previous doing, only faster. It helps people cut corners.
I use it for clean up. I get my image to 95% complete and feed it to the AI, turn the creativity meter to zero, so that it isn't stealing other people's work, and ask it to redraw the image just as it sees it. In the act of redrawing a piece of digital art it will remove all the things it feels are mistakes; pixelation, the white halo around the edges, it will balance the values and contrast levels, it will even out the resolutions of all the layers so they match. it saves me hours of clean up, and the clean up phase is mostly just tedious uncreative labour. This is what AI is good for. It is helping artists cut through the tedious parts of their job so they can focus on the creative parts more.
I think all digital artists better get used to AI cause they'll all be using it for clean up soon. But these public AI web sites aren't going to be around much longer. Soon AI will be something companies will have to buy privately to be used in house. It is just too energy intensive. In fact, some small studios may decide it is cheaper to pay grunt entry level artists to do the clean up work than it is to pay the energy bill of an AI server.
I'm actually not convinced the models WILL keep getting better. We are trained to think that based on the advances in IT over our lifetimes, but it costs $0.36 PER REQUEST for chatgpt, probably even more than that for images. Additionally, we're really approaching the limits in efficiency that we can get out of hardware. Lastly, the current crop of AI is being propped up by investors that think they're taking a loss to get the tech to a place where it is profitable. I don't think it WILL ever actually be profitable. I think that once the scales fall from the eyes AI will be seen much like crypto: something you can waste a bunch of electricity on, but that doesn't ultimately provide much value, and is ridiculously inefficient.
I hope from actual bottom of the very last fiber of my being that you're right
GODS do I hope you're right.
I don't know where you're getting these ideas of pricing from but you're out by a factor of 100.
@@Candlemancer heard it from a journalist that’s been following AI for years in a podcast
It's far less than that for images. GPT is a huge model you need datacenter hardware to run. I can run Stable Diffusion on a GPU that costs a few hundred bucks and only uses 200-300 watts tops when maxed. So the price of a generation is the price of like 0.0003 kWh of electricity, plus the amortization of very inexpensive hardware.
0:18 Not to mention between the players themselves. I don't know a single MTG fan who doesn't have a favorite artist.
This video is made even better by the way that your bookshelf background looks like an AI rendering.
Thanks for this video. One of my festering pet peeves is what I’ll call the “It’s Bad Anyway” argument. One prominent example is “Sure, Chick-Fil-A’s CEO is homophobic, but the sandwich isn’t that good anyway.” I see tons of well-meaning people making arguments like this, and all it does is cede to proponents of Chick-Fil-A that if the sandwich were immaculate, they would be under no moral obligation to stop giving their cash to the company. And whether the sandwich is tasty is a matter of opinion - lots of people think it’s great, and when people say the sandwich isn’t that good, they think “no, it’s delicious; you’re crazy” and the phobes win.
It is the same with AI-generated art. Lots of people think it looks pretty good, and as you say, it will look pretty perfect shockingly soon. So when people cite the blemishes and the alien-looking faces as the reason it’s bad, proponents of these models are rightly dismissive! It’s a vacuum in critical thought - “I’m supposed to not like this thing, so my reasons for disliking it must be simple, sensory, and visceral.” But it takes a lot more thinking to drill down to why AI generated images are bad.
As long as we are under capitalism, as AI starts to replace human artistry in commercial contexts, humans are disincentivized to become skilled artists, and human artistry in total is under threat of extinction. I think humans will always do great art in community contexts, but real art is gonna be real punk, real DIY, and real offline for a while before we hopefully come to our senses about this generative AI stuff.
Anyway, long comment, but I’m just very glad you brought up why it doesn’t matter how crummy the details in AI content look.
Just an FYI - there are plenty of artists who work with 3D as a base, and integrate painting and textures to complete the image. They do this to get crisp lines, correct perspective and lighting. This in turn actually does result in images that are similar in style to the image above. Kinda like a photo, but not. So - that on it's own - not a good indication (obviously the other obvious tells... that's a better method).
Cliff hanger! I really want to hear more about your statement at the end that you don't think AI generators will ultimately be a good thing. Please elaborate in a future video! Thanks Hank!
We didn't even have copyright laws set up properly for creatives on the internet, introducing AI is a mess. Creative commons licensing doesn't have a "okay for use in a data set" permission lol. (not that it was even prevalent enough to be effective at protecting people's work quite yet.)
This is gonna be like when the internet was first starting to be in every household, we are on the forefront of the next big thing... For better or for worse.
This is one of my arguments on the failure of schools and people taking their education seriously.
We are losing critical thought more than should be acceptable. “Fake news” or duped images or even dubious actors can run riot because few are able to discern, or even want to discern, what might be truthful on the internet.
Except that, in aggregate, older people seem worse at identifying misinformation than young adults.
@@jonathancassels4835 I’m not entirely sure it’s a generational difference, really. Education is meant to be lifelong, and there shouldn’t be a major regression in older adults yet I agree that many fall for misinfo often as well.
I think *everybody* is getting pretty crap at this.
@@box3ythe oldest and the youngest seem to be the most susceptible, but I do think Gen X and older are by far the worst about misinfo, especially boomers who just eat up anything they hear as long as it fits with their preconceived notion of the world
This has been a problem for a long time - the subject that has been deemed to be the place where you should be taught critical thinking is English, a subject that's awful at teaching that because its curriculum is set by geriatric shakespeare fanboys who clearly revel in designing the most irrelevant subject possible. People complain that they'll never use trigonometry after high school because they're paying enough attention to know that they're learning trigonometry. People don't even bother complaining that they're spending four hours a week being told about the symbolism in a book about disabled vagrants in a century they've never lived in and a country they'll never visit because English is a class you can pass just by going through the motions - at least in maths you do have to learn how to do the irrelevant equations, not just rote memorise the answers.
I think probably the way to start teaching critical thinking is to replace some (or maybe all) of English with more of a debate and presentation sort of subject, where it's open-ended enough that students are allowed to find topics that matter to them instead of just doing boring old people books, and they get to be motivated by the fun of arguing with people.
gahhhh, you see this can be considered problematic because people who do believe in fake news believe that everyone else in the world is doesn't have the critical thinking discern 'real news'. Thus, continuing the cycle of fake news'.
Wow that AI background behind Hank looks great 😃👍.
😂🤣😂🤣😂
We use other AI tools to produce many products, so I don't see why using it for the artistic aspects is a problem. The only problem is when people lie about using it or when it is used in bad faith. But using it in bad faith is not that much different from how we have used Photoshop to give people self-image problems by making people look prettier for decades. We just need a law that makes it easier to punish people if they knowingly lie when asked if it is fake. Not just for AI, but manually faked stuff too.
I love how Hank's post chemo hair keeps getting darker and curlier
I was at the bank this morning, and they had a giant printed ad on the wall, maybe three feet by two feet, that had an OBVIOUSLY AI generated snow globe image. It was a bit jarring, really
As an artist, AI art makes me feel a bit sad and defeated. But I’m more inclined to understand people who feel that they can’t afford human-made art. What really makes me mad is AI art scams. I’m already seeing AI images used to sell fake products online. On Etsy, we already had a problem with doctored photos for things like house plants, vegetables, and flowering plants. Now they’re using AI to make pictures of plants that aren’t even kind of real. It pains me that people are getting scammed on such a broad scale. My tip: don’t buy plants or seeds online unless it’s from a long-time operating seed distributor or you have enough plant experience to know what’s real and what isn’t. If you aren’t sure, go to a local plant nursery instead
I am really hoping that humans will have the same reaction to AI as they did CGI, ie "This is really cool" to "ahhh! too much, too much!" (eg- Green Lantern, the Hulk) and maybe the best we can hope for is a blending of human and AI, like the best movies do with CGI. I personally detest the notion that artists are losing work. I recently heard two Beyonce songs as an example, one was AI generated, one was not. I knew immediately which one was the real one. The panel, who were AI experts, all picked the AI song as "real". I shook my head in sadness.
I think the only way you could really certify that a production doesn't involve any generative AI would be to obtain proof of work, like undo history and layers, or physical artifacts, from the entire production team.
Also hand cams. Speedrunners use them regularly as proof that their footage isn't cropped, so maybe the same could be said for art?
The real red flag isn't even the details of the image, it's Hasbro's reaction being, "it was totally a human artist" instead of "This piece is by Jane Smith. Here's a link to more of her work."
Hasbro's response was likely also written by AI. "Of course we're human!"
why would she deserve free exposure. the entitlement of the 'artistic' community is mind boggling.
I must say, your content is always enjoyable. Your insight, your youness, and your articulation make for something that I always vibe with. ❤
What a lot of people aren't understanding is that it's not a dichotomy of "real artist or AI". AI is now a tool that real artists have in their toolbox, which they will use to create real art.
I'll be honest, it feels more like a crutch than an actual tool for an artist trying to improve. It's not that bothersome to find references, and it's easier to take a picture yourself for reference than try and prompt the machine.
AI would work better in roles that don't include generation, in my opinion. Like for you to sketch a rough of a pose, and it finds the references for you instead of mangling them together in an inconsistent mess, that'd be pretty nice. As it stands, it still is real artist or ai, since it's how it's being used to shit out slop. I can't think of any other decent use for generative ai content.
It's not just a "tool" when you know that everything you create now will be accused of being AI even when you did it yourself, that is exhausting. A tool is not something that kills the very demand for what it creates.
@@cosmicllama6910 Eventually people will stop panicking about it and artists will be able to use it to create things they never could before. Didn't you see Corridor's recent videos where they invented a brand new animation technique using AI?
as a "real" artist, agreed. i've seen friends use ai in interesting ways, usually incorporating it with their manually drawn art, to create beautiful works, and they get harassed for it every time. sure, there are people who use generative ai to be uncreative or scammy, but there has always been art that is uncreative and scammy. the problem is how the tool is used, not the tool itself.
Hey! I'm a digital artist who's done illustration and 3D. These are my tips for spotting AI -
- Weird perspective. Check the parallel lines and how objects line up in a room/space. Even images that are cluttered and disheveled fall within the order of perspective. Windows especially can be dead giveaways
- Asymmetry in objects that should be symmetrical
- Similar objects that come in different styles and sizing, when objects in that space should be uniform. For example, putting a simple streetlight next to a more ornate one, but they're a similar size and color. It's odd to see 3 different types of streetlights on the same corner
- Hands and eyes are often a dead giveaway. Sometimes teeth. Hair will oddly blend into itself or will follow a seam in clothing or jewelry
- Light sources don't add up
- Typically has the Pixar studio lighting. Often will not make heavy contrast or shadowy images. Bland mood lighting. More often than not they'll have warm lighting
- Tangent lines. A lot of corners and lines run into each other. It takes away clarity from the image
- Tubes and lines blend into each other in bunches or will follow a seam
But most egregiously, it lacks thought and intention.
That being said, 3D has had procedurally generated stuff for a long time to make the process more efficient, especially as timelines get tighter and tighter. The issue is subtracting the people from the process for something that's "good enough" at a glance.
Casting my vote for "Yes please talk about this more it is vitally important for general audiences to understand deeper because surface level understanding of AI images makes it sound way more ethical than it actually is" please and thank you.
I REALLY hope Brennan Lee Mulligan and some of Hank's other Dimension 20 cast members/ alumni do something for P4A! It was AMAZING to watch Sam & Brennan react to Hank using the Banana Loca with Dr Pepper Beans. 😂 Especially because Brennan is also very chaotic about food. Lol
Hank, you are absolutely the person who can and should make that video about why AI art generation is kind of bad for the future. I can think of very few people on the internet that could both entertainingly and quickly explain the problem and HOLY CRAP WE REALLY NEED A VIDEO LIKE THAT SUPER SOON. PEOPLE DO NOT GET THE PROBLEM!
As an artist, I am greatly interested in where this goes. I feel like it could be a valuable tool early in the creative process, but then should never be used as a final product. It's so random, and the results cannot currently be copyrighted. The fact that the source data was used without permission, or without compensating those whose names get used as prompts, is also concerning.
The fact generated imagery can't be protected by copyright is actually a point in favour of using generative AI, not against it. Copyright is the ultimate enemy of art and of artists. The more products are made using AI and therefore uncopyrightable, the more room the world will have for art.
@@yurisei6732 Lol what, such a bad take
@@yurisei6732 this is like saying "the more fast-food restaurants we have, expanding over traditional home made ones, they better
Cause then people will have more food"
Meanwhile "food" is a fucking big mc
@@yurisei6732 The world already has that art. Were do you think the AI ripped it from?
@@yurisei6732 Copyright is not the enemy of art and artists. Excessive copyright that protects corporate interests over creativity is. Copyright is a necessary protection for people to be able to make a living off their creations. Without it, you can spend a week creating a beautiful work of art, and I can come along, copy it, and sell it for cheaper because I don't have your overhead.
Copyright is only a problem when it lasts a century and prevents anyone from doing anything with any aspect of the creation. There should be two kinds - derivative copyright, and reproduction copyright. A reproduction copyright should last the length of the artist's life. No one can republish the art in whole (either singularly or aggregated). A derivative copyright should last 10-20 years, max, and should protect against someone creating derivative works - either using your characters, or creating an adaptation, or merchandizing, etc.
What is the ultimate enemy of art is corporatization. It's art made for the mass market, with focus groups and market research designed to produce something that appeals to the lowest common denominator. Which is exactly what AI art will produce. Just ignoring the fact that it is stealing other people's works and putting people out of work, AI creates bland, soulless, aggressively average fluff with no style or substance. And when that's all that corporations will produce because it's the most profitable, that's the true enemy of art and artists.
I can't remember the specifics, but it was a few months ago at least when there was a court case in the US over claiming copyright on an Ai generated image, ... It turns out the precedent is now set in law that AI generated media cannot be copyrighted because it was not created by a person. This could potentially make image creation a lot cheaper for corporate entities, but it would take away any control they have over the use of the image. It could also potentially put a lot of artists out of work completely, much the same way computer coloring in comic books put all the hand colorists out of work by the end of 2003 when Chaos comics finally shut down.
I work as a graphic designer and sometimes use a graphic website that I pay for to save on time. In spring of last year I started noticing tons of ai images on the website, all of them were jpgs, which was extremely annoying. By fall the website has moved them to their own tab, but they are still there. I can see how easily someone could use AI images and not even know.
With every video Hank posts, he looks more and more energetic and healthier 😊 hope you’re doing well!
The reason that I felt bothered by this situation with Hasbro is that they had just laid off 1100 employees just last month. I only started playing this game just this past summer, but I have already seen a huge decline in the quality control of their cards. I will open packs and have curled cards, misprints, ink issues fresh out of the package. Now after firing a bunch of their staff, we get this situation where they are claiming that they didn't use AI for artwork until they were proved wrong. I wonder about the next time they will lie about cutting corners on something else...
In 6 months that's not decline, that's just the standard. MTG has been like this quality-wise for years. Also, far from the first time WOTC has been shitty, and it'll be far from the last. You're going to have to figure out how to make peace between that and your continued enjoyment of the game - I personally recommend switching to exclusively using proxies, this is the most reasonable middle ground between being able to play MTG and not supporting WOTC very much.
The enshitification of everything marches on
I'm not entirely convinced that the background of this image isn't AI in some kind of elaborate Hank Green troll!
My go-to method for how to determine whether I should scrutinize something for signs of AI is the same one I use for scrutinizing for misinformation: How I feel when I look at the image. If it provokes a big strong "Wow!" or "What???!" or "How!!!?" I immediately take a step back and consider why I had that reaction and whether the creator of the image was attempting to provoke that exact response.
A few months ago, I saw a photo of a woman in a mermaid costume swimming in an underwater cave. I had that big "Wow!!!!" reaction and immediately became skeptical. But I looked at the picture closely and read the accompanying description, which gave detailed information about how the shot was taken and how it was a high-risk shoot with a literal olympic swimmer and they only had one opportunity to get the shot right. That amount of highly skilled work actually lined up with the "Wow!" response I was having, and the fact that there was only one photo, not a collection of 20 pictures with other mermaids in similar scenes, confirmed for me that it was real.
There's a lot going on around the art application of the technology and it is disappointing that even in these early stages it is being used to take bread off people's tables. Wizards of the Coast specifically did a bit of a yikes when it dropped a chunk of its art department while or just before hiring for artists who specialize in touching up art. Which really brings in the 'Where do we draw the line between human and AI created art when AI tools are involved?"-
MtG has always been excellent in at least one thing for the artists: It credits them. This is more than most organizations do, especially as we've moved out of the 90s where art could have identifiable signatures in at least the games and book publishing business. In an unsettling way this is speedrunning a course mapped by the industrialization of textile manufacturing and 'fashion'.
AI can be used for so many things. Why are we wasting it on things that are unnecessary and humans already enjoy doing
Because there is profit to be made unfortunately, Capitalism wants the advantage in every scenario.
I really appreciate that you call it ai generative images. I've been calling it ai gen for short.
Hire artists, dont rely on AI. People use techniques and experience to convey a message in their work, regardless of what you think or want (besides for commissions and even then the artist still has a ton of freedom to convey whatever they want ie. The Mona Lisa). AI copies those people to give you what it thinks you want. There's a big difference and there always will be.
And yet people are increasingly unable or unwilling to tell the difference, proving that whether or not the difference is there, the difference doesn't matter - especially for something like MTG which has such a strong corporate style that there's no freedom or artistry in it anyway.
@@yurisei6732 " a strong corporate style that there's no freedom or artistry in it anyway"
This hear tells me you have no clue what you're on about
@@Americanbadashh You've never seen an MTG card have you?
AI art has taught us a curious new quality of the uncanny valley: it can become wider. AI art can get as good as it wants, it might eventually figure out hands and text. But as it improves, and we see more of it all over social media and image searches, we'll recognize new signs. The uncanny valley, by definition, is always *JUST* before an artificial rendition looks like the genuine article, enough so that its artificiality is obvious and unsettling. But as Hank alluded to, it remains to be seen if anyone will care, or notice.
Hank PLEASE make a video about image generators and why they’re terrible for content creation in general, it’s especially relevant to UA-camrs now considering the Google LLM shenanigans. A full in-depth discussion on the use of AI, both in the context of morality and practical longevity, is direly needed rn
we're long past the point of spotting good photoshop, so in a way it stopped mattering if the photo was photoshopped or not. We're pretty much at the point where the best synthography is indistinguishable from other art forms if you want it to be. what a crazy far flung future we're living in! :]
The proof of this is in how meme templates have become increasingly low effort: "fake" content is so easy to do now, and so easy to make look good, that deliberately making your work seem amateur gives it the sort of authenticity vital for a meme.
I don't know who "we" is supposed to be. Perhaps speak for yourself going forward.
@@bleh329AI art is superior.
@@bleh329 nitpicking about wording doesn't make those two situations not true :]
@@bleh329 "we" in this case refers to all of us, that is, humanity
I'm really sorry Hank this was extremely interesting, but my brain kept asking why you have a can of Heinz Baked Beans on the shelf behind you, how old is it? is it empty? is it a prop? Did I miss some deep Vlogbrothers lore? My god what I would give to just look through every interesting little thing you have on your office shelfs.
I'm not entirely sure what the origin of the Heinz Baked Beans is, but I do know it involves a furby covered in them, of which many memes were made in previous P4As.
@pawz3016 has it correct. Heinz Baked Beans became much more of a thing in the last few Project for Awesomes and have featured a furby covered in them (Beanie Sandfurbs), and Hank doing an increasing number of... questionable? Bizarre? Too awesome to be contained? too terrifying to be contained? food challenges with them during P4A livestreams. (A banana loca is involved.)
Somebody should make an AI that detects AI images
People have tried doing that with Chat GPT and I've seen a loooooot of people who did not in fact use chat gpt but still got bonked on the head real hard by their teachers for "cheating".
As hank said, right now it will work, it might even work really really well, but once AI gets better and better and it becomes harder to detect real works that are just quirky in some way will get caught in the cross fire and for what? To catch a company doing something everyone is already pretty sure they're going to start doing?
What happens when it starts flagging non-ai art?
We have that. It works pretty good. No one cares.
@@johnblunt6693 This, too, has happened.
yeah the problem with this is that if you can run any detection algorithm, including an AI one, you can very easily train a second AI to avoid it. In fact this is one of the main strategies for making AI more accurate 😅
But as a side effect, it means pretty categorically you cannot create an AI nor computer program that detects AI generated images or text.
You can make one that works today, with today's models, but if it gains and popularity it will very quickly become outdated & 100% useless. And given that the AI companies basically do this themselves to train image generation AI, it's very unlikely anyone will be able to do even that.
We had voice actor guest speakers come to my Woman in Animation club at college that talked about embracing AI. I vaguely remember one of the men saying that I know its a knee jerk reaction to instantly be angry about anything AI at all, but that we should all really learn to use it to help us in the future. It rubbed me the wrong way a little bit, but I think I understand the core of what he and my professors briefly started talking about before I graduated. Ai should not be used to replace the hard work it takes real human artist to make the things they do, but assist them and become another tool in their toolbelt the way digital programs/tools did when they first came around (I assume a lot of artists had gripes with things like photoshop in the beginning)
That being said, seeing all the companies choose AI art, artists losing their jobs, and insufferable people on the internet claiming theyre artists when they input a few commands to generate an image pisses me the hell off. And respectfully, older artists like my professors have never had to deal with this exact scenario. So while in an ideal world i would love for AI to be used helpfully, I just dont think it will happen until we iron out everything ethically and legally. I was glad to see the writers strike resulting in one of the first demands for AI regulation, heres hoping the animation guild can achieve the same thing this summer :)
Please, I’m desperate for leaders to talk about this. The hopelessness is getting unbearable. It’s so good to see you looking so well though ❤❤❤
finding out hank is a magic fan is a delightful surprise
Gonna be honest, majoring in painting out of passion only to see ai art everywhere and having to get a job doing what i hate has made life living hell.
I imagine there are people in the world that could scrutinize artwork all day to see if it was AI-generated or not without pay. Or even companies hiring professionals to do just that to check for infringing artifacts.
Dope to learn Hank plays/played MtG. Wonder what deck was/is his favorite
I would love for you guys to bring the AI Image debacle to light, there are a lot of artists hurt financially by what these Ai companies are doing with these Models.
I feel so bad for artists who got their art style jacked by AI. So many artists are being accused of using AI when they’ve been drawing in that style for years.
I want to be freaking out about AI because I tend to do that in general, but the idea of Magic The Gathering still being around and relevant to people is just too wholesome in a way ;)
A common reaction from people that played in the 90s and then moved on from the game is being surprised to discover it still exists, like a baby who closes their eyes and assumes that everyone vanished. Magic is bigger than it has ever been.
It's definitely not what is used to be, that's for sure
@@christopherlundgren1700 Aw that's so cool! I hope it's big in a cool way and not in a "stupid corporate managed" way. I haven't played it myself, it was just around when I was in high school and I always noticed the elaborate artwork on the cards.
Not only is it still around but I'm pretty sure it was more popular in the 2008-2016 era than it was in the nineties. Personally I played a bit in Boy Scouts around 2015 but only got really into it last Autumn. The game is fun but there is a general impression that Hasbro is pushing WotC to make the game ever more profitable since their stock is slumping and Magic is one of their most profitable IPs. This leads a lot of products that feel like they're overpriced for what you're actually getting.
I play it, and my little brother does too (17 and 13)
Hey Hank! I was wondering if AI art reaches a critical mass in our cultural sphere, since the vast amount of image training data is currently pulled from internet scraping bots, is it possible that AI will get worse over time and not better once it starts mostly training off itself?
I personally think it will work like genetics, where the repeated flaws in the system compound, producing even more images with these flaws, creating more flawed training data, ad infinitum. Seems to be a big problem with data aggregation overall, and we can see the effects with websites like EDH Rec
I think this will definitely be the case very quickly if lawsuits can prove they have been training on copyrighted works without the right holders permission. It will be slower to happen and maybe less extreme if they continue to be able to train on new human generated materials.
That's when model collapse stars babyyyy
This is purely a specification problem in AI that can be fixed.
With how AI are currently trained, yes, this is an issue. However this is not innate to how an AI like this HAS to be trained. Improvements in the technology can change this.
For example, advancements in few shot learning may make it possible to train the AI using more direct human feedback or fewer selected hand picked images rather than massive databases.
Also, better specification of what is actually desired of the AI can also improve performance. Right now AI are trained to produce images "in the style of" or "roughly similar to" images in its database, and then matching prompts to images in that database. This is why AI art tends to look kinda samey.
However, an Ai art program trained on purely human feedback or a better specification of "make art a human would enjoy" or "make good art" if we can properly specify that better might be different.
The issue is our best proxy for "make good art" is to basically judge its performance relative to examples of good art. Thats the current approach, and why massive databases of art are needed. But there are different ways to train AI that dont require this approach. This is just the approach that worked first.
Honestly, in 10 to 20 years, AI art probably wont work the same way as it does now. Advancements in training methods, better specifications of complex problems, and advances in stuff like GANs or human assisted stuff means that how theses AI are actually trained can change drastically, and which much improved performance closer to what we ACTUALLY want it to do.
So yes, what you say is an issue, but only in the short term with how these current generative AIs are currently trained. This will likely not be as much of a problem in a decade or two as advancements in the technology are made.
If the quality drops make an ai that filters out "bad" training data and resume. It will get harder over time but any fantasy of ai art going into a self destructive cycle relies on its makers doing nothing. They wont do nothing.
I hope you do make a video about generative AI, because I think people could benefit from learning how it works at more than a superficial level. Common Crawl data, LAION, and also the legal precedent for "fair use" such as Google Books and Google Image Search are super important to understanding generative AI. It seems pretty clear to me the courts will determine it is fair use based on precedent, unless they carve out a major exception for it, which will likely settle the "Is it legal?" debate. Which will leave the "How do we ensure it is used ethically?" debate, because obviously legal and ethical are rarely synonymous.
Ooh, I like the curly hair!! (And would be totally interested in watching a longer Hankschannel video about AI in the future if you were to make one)