The REAL fight over AI art

Ready? And... Think of something that you want to see, something that doesn't exist, like a portrait of King Louis IV as a beagle. The tool I'm using isn't going to go find pictures online and then collage them together. It's going to create a brand new image. You've probably heard of this thing. It's a form of artificial intelligence called Dolly. Dolly. It's called Dolly 2. Image-generating artificial intelligence. These tools are really fun to play with. For the last couple of weeks since we got access, my friend Justin and I have been sending our creations back and forth to each other. I was like, robot samurais. So I put this in here. These are beautiful. With these, anyone can make art in seconds. And artists, understandably, have concerns about that. This technology has quickly become very controversial. We're watching the death of artistry unfold right before our eyes. Justin is a designer and an animator who works with me on Huge If True. He's an artist. I'm not so much. What I noticed as Justin and I were making these is his were way better than mine. It didn't feel as though the AI was leveling the playing field between us. It felt like I was getting new skills and he was getting superpowers. So to test that, we set up a little competition. I think we should try and do a city skyline without AI first. And then we should do like the AI version. And I wanted to get you involved. So we mixed them all up. Mine and Justin's and AI and no AI. And we let you vote on them. 3,000 of you voted and the results were fascinating. This isn't gonna be a story of how AI is gonna replace artists. It's a story like many, many before of what happens when you give people a powerful new tool of how technology, if we use it right, can unlock more of our very human creativity. It's easy to get obsessed. Like when you're using these tools, there's this incredible giddy feeling. Like, oh my God, I can do what? No way. Completely original works of art. This is freaking incredible. And AI just created that image. Holy crap. You know what it feels like? It feels like that scene in Harry Potter where Hagrid's like, you're a wizard, Harry. And there are ways to be better and worse at actually using this magic. Because there's all kinds of like, I wouldn't call them like secret prompts, but it's stuff that most people don't use or don't know how to use. It's Leviosa. Not Leviosa. Okay, let's say you put in astronaut riding a skateboard. It'll give you something like this, which looks pretty cool. But you can get way more specific. You can say a photo of a yellow astronaut riding a skateboard. Now your results look something like this. And let me remind you, a computer is doing this with just the words that I'm giving it. Like, this is already amazing. But you can also erase things and add something else, like with the head of a duck. Or you can go back to the original prompt and replace this part with a blue banana riding a banana skateboard. Sure, why not? Now let's say you like the style of Van Gogh. You could say a banana riding a skateboard painted by Van Gogh. As you get farther into this, you discover new spells. This guy who's just part of the community, like created this Excel spreadsheet, but you can put in like your main idea, like your sub ideas, do not include, aspect ratio, what kind of camera, it's like lighting, colors. So basically he gives you all these options and at the end it spits out this prompt. For example, a poster of a banana riding a skateboard, high definition, bright lighting, vibrant colors, 1950 to 1950 pop art. I mean, like, look how cool these are. Anyway, you can see how fun this is. So Justin and I wanted to test what this could help us do at our very different artistic skill levels. I think we should do some kind of prompt where we challenge each other. What should we do? Here's what we decided on. Round one, make a digital art piece of the New York City skyline, no AI. Round two, make another one with AI. And the only other rule is you can't spend more than 30 minutes on each. Okay, here we go. But real quick, before we get to the competition, if you've watched my channel, you've seen Storyblocks footage. I use them all the time. They're awesome. And they're sponsoring this video. Every time I want great footage, I go to storyblocks.com and I type in what I want to find like city skyline. And ta-da, all of these clips are royalty free and they have a bunch of different pricing plans. I use the unlimited all access plan, which is great. And it lets me try out a bunch of different options. If you're a person who makes videos, I think you'll really like Storyblocks. So go check them out. There's a link in my description. Now, back to the competition. This is terrible. Later. I hate it. I hate it so much. Well, the point is I'm not an artist, so. Hello. You ready? Sort of. I'm ready to see yours. I'll go first with my round one. I felt very limited in terms of my ambition by the things that I knew I could potentially make or not make. And then I found this piece, which I really loved. So I was like, okay, I'm going to take inspiration from this and I'm going to do it myself. So then I looked at a bunch of images to find that tree bottom section, but then I had this concern of like, okay, if I use this, is it, how much of it is mine, et cetera, et cetera. So I went back and I found an image that I had taken myself of a skyline of New York, which is this one. So then I borrowed an image of the sky and that's my final. That's great. You're being very kind. This was actually really good. Thank you for saying that. All right, your turn. All right. This is the final image for one of them. And for this, like I set up a whole set of parameters. The photo gets filled in with a collage image that I put together. And so it's really, really cool. It looks really, really intricate and detailed, but it took me all of five to 10 minutes to set this up. In a way, I'm almost cheating, like the computer would cheat, but it just took me a little bit longer to like know the software and how to get it to do what I wanted it to do. But the results are really, really cool. Justin and I didn't use AI in this round, but we both use technologies that were created to make it easier for people to bring their creative visions to life, which is what the people before that did and the people before that and the people before that. You see where I'm going with this. The AI tools that we're about to use are a new part of that old history. Now, take a closer look at our results. Without AI, Justin and I were both making collages. We were taking parts of other people's creations and creating something new ourselves. People are allowed to do that within constraints. The line on what's okay to use isn't always super clear. Like I felt mixed about using someone else's image of trees, so I ended up using my own. It's messy. This kind of collaging is what a lot of people think the AI tools are doing. Like an image of a banana riding a skateboard must be an image of a banana and an image of a skateboard. That's what Justin and I did, not what the AI does. To really understand why these tools have become so controversial, you need to understand how they actually work. Yeah, great to meet both of you. That's Aditya Ramesh. He helped build one of the most famous of these tools, DALI 2. So DALI is made up of two kinds of technologies, right? Clip and unclip? Yeah. Could you explain how those work? So clip is a model that can be trained on a large data set of like images and captions. If you imagine a Venn diagram where there's one circle for all of the information that's in images and another circle that's all of the information that's in language, it learns the part in between. It's in the process of learning what's at the intersection of those two circles. It learns things like aesthetics, style, and broadly what anyone might write a caption for an image. And so what DALI 2 does is first, before generating an image, it kind of generates Clip's conception of what should be in the image. Then unclip takes that conception and the original text prompt and noise and generates options for what an image might look like. And... Now let's see what we got for the AI-assisted art. Okay, here I go. So first I went with like New York skyline made of mirrors and text out. Like I was trying to get it to do a better version of what I had previously done. This is beautiful. I really like this, but then I had this moment where I was like, I don't have to be limited by what I could do in Photoshop previously. So I tried a stained glass window. I was like, oh, I love this. This is the one I like best of the stained glass window approach. And then I was like, I have 10 minutes left, so let's do it. Oh, here we go. It's like New York City skyline painted in the style of Botticelli. Amazing. See, you're an artist. With a little help. All right, let me show you my round two here. I first started with like Lego brick and then I went with New York City skyline as bananas. What if a kid drew it with crayons? And then I was like, let's now go a little bit more fun. Studio Ghibli, Leonardo da Vinci. I really, really love this one. Then these next few are all by Van Gogh if he was to paint the skyline. So unlike in round one, what we're doing with Botticelli and Van Gogh isn't taking chunks of their art and combining them, but we're clearly using something of theirs. When I say something like in the style of Van Gogh, where does it get the understanding of what is in the style of Van Gogh? So clip is trained pretty broadly on a large data set of images kind of collected from the internet. Most of the time when you see an image on the internet, there's some kind of alt text or caption associated with it. What Dali is doing is less like collaging, that's what Justin and I did, and more like what I was doing when I found inspiration for my original round one. It's retaining information about the style and the colors and all of that and using it later. Honestly, I'm not actually sure compared to what we let humans do, that these AI tools owe more or less to the original artist. The reason we need to be careful about the deployment of this technology is just the scale of it, right? If a human goes out and studies something and gets a source of inspiration from what they see online, that's regarded as acceptable. But now when a machine can do that and basically bring the cost down to something very small. If a machine doing something can bring the cost of doing it down really low, then what happens to the people who are doing it now? That's the real question behind this controversy. Take a look at this. On the left, those are images that were made by a designer at the media company Axios to pair with their articles. And on the right, that was made by prompting Dali. You still need someone to do the prompting, come up with the idea, make sure that it looks good, but the risk, and I don't know if this will happen, but it could, is that because these systems are so fast, that could be one person and not 20. And the chance of that job loss happening in an industry that's already hard is creating real backlash against this technology. A journalist used Midjourney in an Atlantic piece recently and the response was strong. The uncomfortable question that I keep wondering is, is this any worse than what I've benefited from? My job is a video journalist. It would be impossible without the tools that mimic the work of typesetters and in-betweeners and film loaders and a bunch of jobs that I probably don't even know and then put millions of them out of work in the previous form of their jobs. Be honest, is that true for you too? It certainly applies to the software tools used by digital designers in the jobs that we're now talking about. And then what about those who can't afford to hire someone or artists who can't afford to use the current tools? Creativity is distributed worldwide. Access to an art degree is not. Access to the expensive software tools that digital artists now use is not. This is the profound power of technology. When used correctly, it can unlock abundance from scarcity. And in this case, that abundance is human creativity. That's really cool. So right now we're at Dolly 2. Where do you see Dolly 10? I'm more worried about 2.1 for the short film. I am very excited about moving beyond just images. For example, like, you know, audio and video. Things aren't quite at the point where the images now are, like, actually photorealistic. But it's a matter of, like, months before that's pretty much solved for images. Hold on. Months? We need to get ready. We've been talking about art. Photorealism puts this in a whole nother realm, for good and for bad. We're already dealing with deepfakes and misinformation and all of the things that happen when it's hard to tell what's a real and what's a fake image. Like, this is... But here's the thing. We don't need to blindly accept technology as all or nothing. These AI tools are a huge leap forward, pushing us creatively. But they'll need our help to refine how they get used. That's actually one of the reasons I make this show in the first place. I want to help people see these tools early and be a part of that conversation. I don't know how we're going to choose a winner here because I think both of ours crushed it. A friend of mine had the idea that in order to raise the stakes, we should allow people to publicly vote on them. Mmm. Then we just say, like, hey, go to this website, vote for your favorites, and we just see which ones rise to the top. We did do that. 3,000 of you voted. And the ones that you liked best were Justin with AI, then Justin without AI, then me with AI, and then all the way at the bottom, me without AI. Ooh. It's cool. We're cool. I'm fine. I think that we saw in the competition what we expected to see, which was that it boosted both of us. And by boosting both of us, it gave me new skills and it gave you superpowers. Increasing access to powerful creative tools to more and more people has given us so much, has allowed us so many opportunities to flourish. In this story, a group of humans created a tool that potentially billions of other humans could use to make the ideas in their head real, faster and at a fraction of the cost. The possibilities are endless. I think as we've created new technologies, whether it's, like, new softwares or, like, new ways of creating art, like, we've gradually gotten more and more and more inclusive. And, like, this kind of leaps us into this other realm of, like, anybody can create art and it can be really cool art. So, what will you make?