--:-- Host: On June 27th, our team headed to New York City. We are at the A16Z office for the first ever AI artist retreat.
--:-- Justine Moore: That was A16Z consumer partner, Justine Moore. Justine was one of many partners who attended this retreat, which brought together the builders behind some of the most popular AI creative tools in existence. That is Eleven Labs, CREA, FIGL, UDEO, Ideogram, and Civit. All together with 15 top artists. These are the folks who are often doing the coolest things with these sorts of tools. They're kind of pushing the boundaries of what the tools can create.
--:-- Host: Today, you'll get to hear from many of these AI founders who, together with these artists, are advancing what it means to be creative.
--:-- Anish Acharya: Art is going to get better than ever. The average art output is going to improve, but so is the ceiling. It also is a higher participation rate. Everyone who's interested in creativity can be creative and express themselves, which is just so cool.
--:-- Host: That was Anish Acharya, general partner on the consumer team. But that's not all. I've been a founder twice. I've been spinning records as a DJ for 25 years, and I'm all about AI and art. So what happens when you put all these investors, leading artists, and creative tool founders all into the same room?
--:-- Justine Moore: I mean, the vibes have been immaculate. And I think that the thing that's most surprising is how much everyone has in common. Like the founders are more creative and the creatives and artists are more technical. I think the other thing has just been how interdisciplinary it all is. People making video want to play with generative audio. People making music want to play with sound effects. It's just incredible to see. One of the coolest things was a lot of the founders had recognized people by their online screen names or knew, "Oh my gosh, you use my tool to create this incredible song that went super viral," or "You use my product to make this kind of amazing video animation that our whole team was talking about for a week." These are people who have been interacting with each other often daily online for the past 6, 12, 18 months, sometimes even two years, but didn't even know what each other looked like in person.
--:-- Host: Now today, you get a behind-the-scenes look into this event, including the origin stories behind many of these tools, which, by the way, some have never been shared publicly, and how these tools, which have all gone through their own viral moments, are navigating this AI wave and what they see on the horizon. Let's get started.
--:-- [Disclaimer: The content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com slash disclosures.]
--:-- Host: Here we are in 2024. We're at an exciting inflection where your creativity is being unbounded by the tools available. I mean, we're early, but there's more people making more art and more people making more tools to make art than ever before. And if you look at the history of technology and art, every single time there's been a new technology, the amount of art has dramatically increased. People were worried that drum machines would compete with drummers. And instead, there's more people making more music with both drummers and drum machines than ever before. So I think there's a sort of equivalent moment here in technology and art where we're at the beginning of everybody who has taste and interest in art being able to make it.
--:-- Anish Acharya: Many have drawn parallels to prior computing waves. But is this any different? Well, what's different is for the first time we're creating these sort of left brain things. Computers and computing platforms have really been in the business of precision. And now we're creating products that are intentionally imprecise, beautifully imprecise. So it just feels like a whole different flavor for products and product design than we've ever seen before.
--:-- Host: So let's introduce you to some of the people behind those products. We have companies here covering basically every sort of creative modality, image, video, music, 3D, speech, all those sorts of things. That includes...
--:-- Conor Durkan: Connor, I'm a co-founder at Udeo.
--:-- Ammaar Reshi: And Omar, I'm the head of design at 11 Labs.
--:-- Host: Both companies are focused on audio, with Udeo focused on music, while 11 Labs is tackling everything from voice to sound effects. Meanwhile, founders like...
--:-- Mohammad Norouzi: Mohamed, I'm the co-founder and CEO at Ideogram.
--:-- Victor Perez: Victor and...
--:-- Diego Rodriguez: Diego, who are the co-founders of Krea.
--:-- Hang Chu: And Hang, I'm working on Viggle.
--:-- Host: These founders are building at the increasingly sophisticated world of 2D imagery and video, plus 3D. Ideogram, for example, lets you generate AI imagery with accurate text embedded, a surprisingly difficult technical feat. Viggle, on the other hand, is building at the intersection of video and 3D. Meanwhile, Krea has come up with a suite of AI tools like upscalers and real-time generation. Or, in the case of Civit, a new breed of marketplace.
--:-- Maxfield Hulker: My name is Max Phil Kolker. I am COO at Civitai and co-founder.
--:-- Justin Maier: Yes, and I'm Justin Maier. I'm the CEO and co-founder of Civit as well. And CPO and CTO. Lots of things. The joys of a startup.
--:-- Maxfield Hulker: We are a massive community of people making tons and tons of AI creations using community-made models with community-made patches to those models called LORAs. We give people the ability to either train on a few specific models. So a model focused on anime or a model focused on being semi-realistic, or they can select their own custom model to train on top of.
--:-- Host: With AI moving so quickly, it's clear that we no longer live in a world of just ChatGPT and Midjourney. Numerous companies have springboarded into the zeitgeist and grown at unprecedented rates. So we thought it was fitting to take a step back and document this whirlwind of a journey. And while many of these founders have been quietly working in research for years, their origin story often started from scratching their own itch.
--:-- Mohammad Norouzi: I guess part of it is that there is this thesis that everybody has an innate desire to create. And as humans, we have this inner creative child. The education system sometimes kills this creative child, unfortunately. And what's finally possible with technology and AI is to help people express themselves visually and creatively. So that's the interesting part. When you think of using image for communication, then you can communicate much more effectively if you have image and text together.
--:-- Host: For Mohamed, it really was this unique combination of text and imagery.
--:-- Mohammad Norouzi: For me, image and video is dear to my heart and very personal.
--:-- Host: But for Connor, it was his connection to music.
--:-- Conor Durkan: Music for me, I think, is this very special medium. It's everywhere at all times. It's in the background when you're at a restaurant or cafe. You're listening to your headphones when you're going to work in the morning. It really has an emotional resonance with people. And for me, making that abundant, the promise of generative modeling is that a lot of this can be far more abundant than it ever was before.
--:-- Host: And for Victor, it was his discovery that programming itself was the creative gateway.
--:-- Victor Perez: With me realizing that you can use artificial intelligence in a creative way.
--:-- Host: But for Ammaar, it was building his own side projects and a desire to share what he was learning that actually propelled him into his role at 11 Labs.
--:-- Ammaar Reshi: It's really funny, actually. Over the last maybe couple of years, I started diving into AI tools when ChatGPT came out and started making things on the side for fun. One of those was a children's book that ended up accidentally going viral. And that kind of was my journey into AI. Through that and making that book, I started exploring other AI tools. And what I really enjoyed was sharing what I was doing and how I made it. So I discovered Eleven Labs and made a podcast with Eleven Labs where I was talking to a fictional figure. And we were having a back and forth conversation that also kind of did the numbers on Twitter. And then I was like, "I love using this tool. I'm going to make my own AI short movie." Actually, Justine and I are friends. And so I showed it to her and I was like, "I kind of need free credits because this movie is using up all the credits on Eleven Labs." And she's like, "You should meet the founder, Maddie." We met, we really hit it off. And Maddie, in classic Maddie fashion, was very direct. And at the end of the call, it was like, "Hey, we're actually looking to hire someone to lead design. Are you interested?" And then to work on a product that I'd used for over a year.
--:-- Host: That experience also gave Ammaar a taste of just how quickly the space moves and also a hit of virality.
--:-- Ammaar Reshi: It happened because a friend of mine had their first kid and I read her children's book actually. And I was reading it and I was like, "This story kind of makes no sense." So I went back home. I'd been using Midjourney a lot, ChatGPT two weeks old, combined the two to create that book. And then I was like, "How do I get this published?" And Amazon has this amazing publishing service. You can get a book out within 48 hours. I had a paperback in my hand in 72 hours. It's so fast. And it's really interesting because writing a book and publishing on Amazon was almost like iterating on software. If I discovered a typo or whatever, I just updated the PDF and the new book was out and a new publishing line was out. So, yeah, I put it out there, got a ton of virality from that. And yeah, that was a really interesting experience.
--:-- Host: Pre-AI, we were in this era of consumer where it was just really hard to get people's attention, really hard to get them to download a new app or try a new tool. You had to spend a lot of money on customer acquisition. Now, just with the real excitement around AI, if you make a cool product, you can get it into the hands of people and get them using and talking about it. This was the case for Victor and Diego at Krea, who eventually met their own viral moment, although it didn't come easy.
--:-- Victor Perez: First of all, it was called Geniverse, coming from Generative Universe, best name ever. And essentially, it was like two things. It was on the one side, an open source library that was kind of integrating all the cool stuff that was available at that moment. And on the other side it was a creative tool. And the way how it looked was super experimental. We didn't really know how to do UI design or any of that. The background really had stars and everything. So it was like a galaxy, like the generative universe, right? And then you could put text, you could put images, and you had a few things that you could tweak, and you could generate images. And you would see the image evolving in real time. And the images that you liked, you could keep them. And they were added to this kind of universe. And essentially you ended up with a ton of images in this interactive space.
--:-- Diego Rodriguez: So for us, it was always with the same idea in mind. On the one side, controllability, and on the other side, intuitiveness. How do we make tools that don't look daunting? Because AI in the end is a new creative medium. A lot of people are using it for the first time and we want them to have an experience where the AI does what you expect it to do. And you don't need to learn about crazy prompt engineering and all these tweaks to get good results. And on the other side is controllability because we are dealing with creatives. We are dealing with folks who are not just okay with having a beautiful image. They want that beautiful image. So these are the two core principles that we had since then. And we built kind of a Figma-ish interface for AI. And we had every single utility that you had at that point with Stable Diffusion in there. We had thousands of AI models that you could use. We had every single technique, every ControlNet, everything was in there. But, you know, it was not working. It was a learning curve that some people were just not willing to take.
--:-- Victor Perez: So then we had the first kind of virality moment when we shipped this thing that was almost like an equivalent to a meme generator. I remember that we were seeing all of these images on Twitter with the spirals, right? And we were like, "Well, what's going on with the spirals? We can do this. This is like one day of work." And I remember at that point, Diego was like, "We should do something with this. It's getting so viral." And I was more in the mood of, "No, we need to ship this other feature." We were working at that moment until at one point we were like, "Okay, let's do it." And we did it in the sketchiest way possible, in one or two days, and we shipped it on Twitter and it went viral. It was the first time that we lived something that I had read about in terms of this is what PMF looks like. That's the first time I was like, "Oh, Jesus Christ. Okay. This is how it looks."
--:-- Diego Rodriguez: Okay, as in you go to sleep and I can feel the heart beating too. And then you sleep three hours, you wake up because you know that there's stuff broken. The email starts to get flooded, Twitter starts going. Suddenly, literally every day was crazier than the one before. I was like, "Oh my god, a thousand people. Oh my god, 10,000 people." And it's like, "Oh my god, Football Club Barcelona, number one soccer club, just used us. What? Why?" Like, how many followers? Oh my god, 100 plus million followers on Instagram. Okay.
--:-- Victor Perez: I feel like it was actually hard in the sense of, as a founder, you're like, "I've put multiple amounts of years into many things. And then the thing that we literally think is not important gives you all the success." So it's a moment of reflection. You're like, sometimes the world throws truth at you. But those years of work were not all for nothing. And I don't think the years that we've been working on was a waste. No, it actually is. Oh, that's what you learn on the technical level, how it works. I mean, because it was so much failure, we learn about, okay, how do you communicate with your co-founder?
--:-- Diego Rodriguez: I think that's something important to note about those times is that we were very, very aware that this was a trend and that this was not the end product that we were building. It was almost like a marketing engine that we were using to get better branding and to get known. They were finding us because of one reason, but they were staying because of another one, which was this other product that we were working on. Even we knew that.
--:-- Victor Perez: I think that the core learning that we got from this experience is that the AI field changes constantly. Every month or every two months, there are new breakthroughs, new techniques, new ways of doing things. And the tool that we were building was already starting to get too complex because we were trying to put everything in a single tool. And I think that what we learned with the experience of the Spiral Virality is that there's a lot of value in simplifying super niche and simple use cases. And that was the case again when LCMs were released, right? We saw this technology and at that moment we used all the experience that we got from the first virality to engineer the second one. And the second one, we knew that it was not a trend. It was something extremely valuable. We were finally able to get that interaction that we were looking for for almost years, right? We can generate images in real time and have full control of the colors, the composition, the shapes, everything. That was almost like a dream come true.
--:-- Host: Victor and Diego have now hit virality several times over. But can you engineer that momentum? In some cases, it's all about having a single critical feature not offered elsewhere.
--:-- Mohammad Norouzi: So basically it was the version 0.1, as we called it, and this was back in September of 2023. And it was a model that was working. It wasn't perfect, but we felt like it was already good enough to give to users. And it was the first model that could put legible text into images. So it kind of went viral because of the unique capability of the model. Somehow the ability to put text into images felt needed.
--:-- Host: But in other cases, it's about cleverly enabling the masses, or in this case, the memesters, by drastically reducing the barrier to participate. Here's Hang with Viggle.
--:-- Host: It went pretty viral, right? What was that like experiencing putting a product in the hands of so many users and also seeing that kind of spread on its own?
--:-- Hang Chu: Yeah, it was. We didn't anticipate that for sure. In the very beginning, we were thinking mostly targeting content creators. But somehow the meme makers, the memesters, they caught on to it. And that's how it got pretty viral. And that's also thanks to some of the templates. We spent so much time discussing why this is the case. There was this template, the Joker, Leo DiCaprio coming on the stage, and there's a Joker character that replaced him in the video. And we've seen millions of different characters just remixing the same moment. And we realized that the main reason was it's easy to use. It's so easy to basically upload one image, and then one click, choose that template. And then in just a matter of seconds, you'll have yourself in basically that same moment.
--:-- Host: Maybe one other aspect of the virality is, as you said, the meme makers got a hold of it. There's this kind of fun, maybe even silly aspect to it. How have you thought about that?
--:-- Hang Chu: Well, I think that speaks to the entertainment value. And for anything to have real entertaining value, it has to work. It has to work well. And that actually requires a lot of rigor on the research side. So we are pretty serious about being silly. And it takes quite a bit of rigorous research to do that. And the second thing is you have to have a tool that provides precise control. And then because people are getting what they want, they can have all kinds of variety of fun with it.
--:-- Host: You've mentioned characters and templates a few times. What are some of your favorite examples of those generated on the platform?
--:-- Hang Chu: One is the Joker coming onto the stage template. That one is basically the moment we realized, well, actually, people want to remix. And there is this variety and memes aspect of it. And the second one is there has been one Rakuten advertising song and people are dancing with this. And we're also seeing millions of people remixing that same template. This is interesting for us because it makes us realize that as long as there is this fun element to it, people actually don't mind this content having a little bit of a brand message.
--:-- Host: When you think about applications, and I know it's early days, but have there been any that have surprised you about the ways that Viggle has been applied? Every time a founder creates a product, they have applications that they envision. And then the best products are often people using them in alternate ways that surprise them.
--:-- Hang Chu: Yeah, that was exactly the case for us. In the very beginning, we were mainly thinking of movie makers, game makers using this as a quick animation, pre-visualization tool for them. It's actually pretty useful for that. And we've also seen the early users adopting to that. But then we never anticipated the meme surge. So since that, we'll be also providing those templates. We've been keeping track of the latest trendy dance moves, sports events, etc. And we've also seen content creators hopping onto this. They are actually reaching out to us saying, "Can you feature our dance or song on your platform? And then can we collaborate on promoting some of those?" That's been really interesting.
--:-- Host: We asked Connor the same question around what he's learning by seeing how the masses are using Udio.
--:-- Conor Durkan: The model we originally launched was a model which generated 32-second clips. And so to make a kind of a full track, you would extend that in various directions. You would add an intro, add maybe a chorus and an outro and stuff. And you would build a song like this and you would start with these chunks. And I suppose we've actually come to realize quite quickly that people's experience with music when they ask for, say, a song is actually a lot more focused than that. So they kind of want a song that begins at the beginning, maybe ends at the end. It doesn't have to be long. It could be like a short two-minute clip, but it has a verse and a chorus and a verse. And there's a structure to it. And so I suppose we actually underestimated just how kind of important that was. And so that's something we were making steps towards rectifying recently.
--:-- Host: 11 Labs was also no stranger to the surprising and inspiring user behavior.
--:-- Ammaar Reshi: Yeah, I think one of the most surprising ones was people who had lost their voices and then had used 11 Labs to, one, bring their voices back to life and then do the thing they love doing. So we had Lori Cohen, who was a lawyer. She lost her voice one morning and a friend of hers helped her replicate her voice with Eleven Labs. And then she was back in the courtroom delivering arguments. And that to me is just such an incredible moment because you don't expect that. And I think our idea was like, "Hey, we're going to give ideas a voice with our product and our tools," but this gave someone their own voice back. And I think that was such an amazing thing to see. And we saw that again with a climate activist, Bill Wheel, who was delivering his award speech. He suffered from ALS unfortunately, but again was able to replicate his voice and then deliver that award speech. So I think those kinds of things are just like, you're like, "Wow, technology is being used in a way we didn't see it." And now we want to lean into that and, of course, help others.
--:-- Host: Maybe in the opposite sense, have there been any applications that you've actually built or designed for where you're like, "Everyone's going to use it for this, obviously," where that's actually not been the case?
--:-- Ammaar Reshi: It's interesting. When we launched dubbing and automated dubbing, we thought, "Yeah, this is it. Everyone's just going to use automated dubbing. It's going to be great." And of course with dubbing, one of the most important things is accuracy, right? And so automated dubbing, we realized people still want a lot of creative control on that. And so we ended up having to build Dubbing Studio, which allowed people to really fine-tune that dub and change a lot of the content. And then we also introduced 11 Studios, which was basically creative teams that help you dub your content with professionals who are really good at that. And so we realized that actually was what people needed more of and not just automate everything and all the things, right? And then it actually picked up again. And this is something even when I was working at Palantir, you learn, which is the temptation to try to automate everything or to use intelligence for everything. But actually, there's so much value in having someone in the middle and still having that human touch to take it to that final step was something we learned with dubbing.
--:-- Host: And as these companies get all this new data, it's not always easy to figure out who they should be catering to. So how do you think about what you build and for who? Your team is everyone, in theory.
--:-- Conor Durkan: I think what we acknowledge is that we probably have different types of users, like distinctly different types of users. At the very top being someone in a studio who's making an album, at the very top level. And then at the other end of the scale is maybe someone on their phone who wants, in a minute, just a funny song to send to their friend. And those are two very different experiences. And kind of similar to the kind of output you can get from just an instrument in general. Someone can have a guitar at home that they play just to have fun from time to time, a totally personal thing, not anything necessarily serious, just a way to express yourself musically. And the same way, a professional can take that same guitar into a studio and make it part of something fantastic. We like the technology to basically enable all ends of all parts of that spectrum.
--:-- Host: Several are unsurprisingly using their flywheel of new users to inform their decisions.
--:-- Mohammad Norouzi: Yeah, we kind of use our user base and the prompts that they enter into the system to decide how to evaluate the quality of the model and what to prioritize. What's interesting is our users used Ideogram to tell us what they want. So they were like, "We want image upload, we want comments, we want more servers." So I guess the good news is we already have this flywheel of users coming and using it. Some are paid, some are free, and that sets the vision for us.
--:-- Host: Hang from Viggle has actually used these new learnings to expand who they're building for.
--:-- Host: How are you thinking about who you now build for? Are you pivoting or adjusting to incorporate these new use cases?
--:-- Hang Chu: So we are broadening our target audience in this sense. We are seeing this as eventually we're going towards the structure of a new type of AI-powered content platform. And the content platform is really important to have all these creators. And those are still the content creators, the artists, the movie makers, the game makers, the game designers. They are the sources for all those new, latest ideas, all those new templates. And then we'll broaden this into content consumers. Basically, Viggle is a new way to consume content. Before AI, it was mainly like, if I like the moment, I will share it, I will like it. But there's a deeper engagement you can have with that moment. I can basically, I love this moment so much that I want to put my own avatar in it. It's almost like in a parallel universe, I want to see how this looks, relive that moment myself. So this is a new kind of content consumption. And that's actually one of the most important aspects. The virality actually comes from all those creative ideas. So for us, it's all about empowering that creative community first, making sure they have what they want. They have the best tool. They have early access to newer features. They have almost private channels. They have almost unlimited access.
--:-- Host: The team at Krea, on the other hand, is more focused than ever on experimentation. And their signal for success?
--:-- Host: When your users are better at using your tool than yourself. How I think about it is that every tool that we launch follows a similar process. It all starts with a hypothesis. That initial hypothesis needs to come from the founder and needs to come from your own intuition. But we are wrong a lot of times in the way we validate these ideas. And when we are wrong, it's through listening to the community, seeing what they do with the tools.
--:-- Victor Perez: A good rule of thumb, something I found that is a good north star to realize when something is good or not, is when your users are better at using your tool than yourself. That has been key to me. With the real-time tool, I was seeing things and I was like, "How the fuck did they create that?" Same thing with the video tool. I was trying to do a demo, trying to showcase cool stuff. I was trying things and I was not getting there. I was looking at Twitter and all the things our users were creating with our product. I couldn't get to that quality. I couldn't get to those results. Every time your users are using your product better than you are, that's a good sign.
--:-- Host: Meanwhile, Justin and Max at Civit are charting new ground, but also figuring out new limits.
--:-- Justin Maier: Stable diffusion allowed you to make anything. When we launched, I wanted to make sure we could continue to support that community. It was so diverse. There's a running meme of things you can make with stable diffusion. In the front is somebody making funny memes, and then there's a train coming that's porn. People know that you can make all of this stuff. That's the point of this tech: make anything. It was important for us to say, "We want to be able to support this tech as it develops." That means we need to embrace all of it. That's not easy. It's been incredibly difficult to set up policies that allow the creation of all things in a way that's not going to hurt people, and to do it in a way that gives people the level of control they need to prevent the creation of content that can't be there.
--:-- In the beginning, our policies were very straightforward. They were kind of just, "Look, as long as it's not illegal, and as long as it's not just ethically completely debased, we'll let it on the platform." We were okay when we had a small enough user group, leaving it even that vague. We found over time that we've had to really specify, because it turns out there are subsections of the internet that are into the absolute strangest things you've never heard of. Some of that is really funny and cool. Some of it is just, "Oh my gosh." It's a balancing act of figuring out, "Okay, what are..." You almost have to grow as a person. We created a council of moderators on our platform to get together and look at when these new things pop up and be like, "How do we feel about this?"
--:-- One of the things that really blew my mind when we were getting into the moderation aspect was, "Oh, we'll just do what other platforms do. We'll do what Imgur does. We'll do what Reddit does. We'll just copy what they're doing." As we dug into what they do, they don't define any of this. None of this is defined. We had to come up with terms. How do you define what a child is? How do you define what is photorealistic? How do you define what is and isn't? All these terms that before didn't really have any set definition.
--:-- Host: Perhaps it shouldn't be surprising that there are new moderation challenges since this industry is so fresh, with new ideas coming from a new breed of creatives. In fact, we heard about this range in both prosumers and professionals from most of the founders we spoke with. Here's Victor from Krea.
--:-- Victor Perez: The range of creatives is quite wide. The kind of people that use Krea can come from having 20 years of working in the creative industry, being a 3D artist, or people doing graphic design, or even photographers. But we also find a lot of folks who don't have a professional creative background. For the professional ones, you can find them doing a lot of prototyping. For example, when they start working on a new project, they may go to Krea to quickly brainstorm some ideas. They use the real-time tool we have for that. They can do a very simple sketch, add a text prompt, and have something that looks super realistic. That can either give them ideas or maybe even serve as a final deliverable depending on what they're doing. When we're talking about less professional creatives, it's honestly more about having fun. They are using Krea for everything you can imagine, from imagining new walls to creating paintings to creating characters, all sorts of things.
--:-- Host: As more participate, these new platforms generate new talent, but also new expectations, like expectations in speed.
--:-- Victor Perez: In the meantime, what we're doing is building community and bringing to the community what they want now. We're just focusing on what can we do now with the technology that is out there. We are very deep into AI communities. Every time there's something that we think is valuable from a creative point of view, we go ahead and execute it very, very fast. The way we're working is almost like a video game company, where instead of video games, we are building tools. Every six months or so, there's a new tool, because the space just happens to evolve in a way that every six months, there's a new technology you can use to make a new tool. That's going to keep being like that until we get to these real-time multi-model systems that allow us to do something way more interesting.
--:-- Host: This new wave has also shifted people's willingness to pay. Back to Anish.
--:-- Anish Acharya: I think the willingness to pay and the amount that consumers are willing to pay is really high. That's really interesting because for so long we've had these sort of patronage models for how to fund the arts. There's been this belief that there's a decreasing interest in paying for art. Instead, we're seeing the exact opposite. People want to pay for art and pay for tools to make art, and pay a lot. That's a really, really exciting development to me.
--:-- Host: And this willingness to pay is also unlocking new business models.
--:-- Justin Maier: People make so many things because it's a tool for creating anything. To see the things that people can create, whether that's assets for a game or videos of flowers that are dancing, it's endless. The possibilities are endless. It's inspiring to see how people are kind of pulling it to do new things.
--:-- Host: That was Justin from Civit, which is also working on a new way to reward AI artists for their contributions.
--:-- Justin Maier: When we were getting this going and we were really realizing this could be a business, we interacted with a lot of the people who were doing this creation. It's a lot of time, a lot of money, and a lot of technical skill that goes into making these things well. People were doing it. Thousands of people were doing it just for the love of the game. They just really enjoyed the clout and the entertainment factor they got. Get their position on the leaderboard. The leaderboard, oh my God, the leaderboard. It became pretty clear that this is almost like a whole new creator economy that can come out of this. It's a group of people who are putting effort and love into something that could very easily become livelihoods for them if they had even the smallest way to monetize it based on the number of eyes they're getting and uses they're getting. A very clear goal from the very beginning was, "Let's figure out how we can help creators monetize while maintaining the open source ethos."
--:-- We actually just announced something that we're hoping to roll out over the next six weeks. Let me give you a little bit of history. We launched a creators program four months ago, and we opened it to a small cohort of essentially 50 creators. We opened applications and took people that met certain criteria and have been experimenting with ways we can help them monetize their work. What we've landed on for this next generation that we're hoping to open up in these next six weeks is making it so that people can earn for the generation that people are doing on our site. If they make a resource that's intended to produce a new character, like a consistent character that they've made, and somebody chooses to use that in the generator, then they're going to get their share of 25% of what we charge for that generation. The aim is to make it so that these people have a way to get essentially paid for allowing the convenience of using their resource on our site.
--:-- One of the main things we saw right away, before we had the time to implement any real monetization stuff for creators, was we put in a DM system simply because we knew that a lot of people are contacting creators for work outside of the platform. Because of that, we get an untold number of people contacting us being like, "Thank you so much for this platform. Because of that, I was able to get hooked up with Hugo Boss or Hyundai or some of these other people who are suddenly using this technology. It's completely changed my life. Before I was making $30,000 a year as a waiter or whatever. Now I'm making six figures doing this whole new thing that's a passion for me." I have lost count of the number of people who contact me about that. It's really cool to see. From a services side, we want to enable that and make it even easier for people to be able to sell their services, their expertise directly to businesses from the system.
--:-- Host: Civit is not alone here. ElevenLabs is also building a marketplace for voices.
--:-- Ammaar Reshi: I know you guys are building kind of a marketplace of sorts so people can upload voices or they can use voices that others have uploaded. Yeah, I think it's a really exciting way to give folks a way to earn passive income as well. Maybe you were a voice actor and you weren't getting the gigs you wanted, but now you can put your voice out there and you might become extremely popular. We've seen people earn quite well on our platform. The library is just a great way to put your content out there. We want to partner with more voice actors, honestly, to have more expressive voices and then give people great voices to create content with. It's a two-way street.
--:-- But it's not just the marketplace, it's also the interface. I think we've always had the dream of voice interactions with all our products. If you think about Star Trek and Knight Rider talking to his car kit, it's something that's been a part of pop culture history forever. But I don't think we've had the quality and the sound for it to feel as natural as it should have been. I think we're getting to that point where the interactions between large language models using voice interfaces is becoming incredibly natural and feels like talking to a person. I do see a future where a lot of this physical interface that you're tapping around with is going to just fade away. You're going to be able to ask the questions you want to ask and have the conversations you want to have.
--:-- I know *Her* is the hot topic movie of the AI space, but there was one thing in that movie that stuck with me more than just the interactions he was having with her. There was a scene in the movie where everyone was down looking at their phones and kind of scrolling. There's this inflection point somewhere later in the movie where actually everyone's kind of talking to something in their ear. I think that is a very prescient take that they had. I think we're going to see more of that. It's just going to be natural conversations we'll be having with this AI or any interface.
--:-- Host: Yeah. It reminds me of my husband's grandmother. The first time she ever heard someone talking on a phone in the grocery store, she thought they were talking to themselves. All of these new interactions, you're just not used to. Or the people who go to prison and come out 10 years later, they're like, "Why is everyone looking down?" They don't realize that we have these crazy computers in our pockets.
--:-- Ammaar Reshi: Totally. The thing that I love about AI in particular and all these AI creative tools is the magic. You had an idea and now you can imagine it. You can imagine the image that you wanted that was in your head and the dream you had. Now we're saying you can imagine the sound that you're probably hearing in your head that no one else can hear yet.
--:-- Host: But it's not just a new UI. Perhaps it's a new approach to modeling the world itself. Hang from Viggle.
--:-- Hang Chu: One thing I really look forward to is the next generation of the model. We are really hoping to extend this character model to more of the rest of the world, like objects and scenes. There are two general paths towards modeling the real world. One is more on the pixel level approach. Diffusion models are really good at that, but it has this drawback: it's really hard to manipulate pixels. The reward is essentially 3D is physical. Pixel is not really an efficient representation for it. But it has the advantage that you can train with any video and generate anything. The hope there is if we scale it up to a certain extent, controllability will kind of emerge. We're taking another kind of different path. We want to nail down ability first, making sure it's just as precise, as controllable as a graphics engine. Then we scale up from there. I think this is how those two paths evolve and how they can be combined into one immersive experience.
--:-- Host: As we close out this episode, it's hard to understate just how much these tools are shifting what it means to be creative, both to existing artists and to those who never would have called themselves artists before. Conor from Udio.
--:-- Conor Durkan: The threshold for someone going into a studio and recording something was way too high. The promise of the technology is that it brings an order of magnitude or two orders of magnitude more people into the creative experience. People can express themselves in this way. More concretely, as moments happen in the world, different cultural moments, you can attach music to them now because it can be dynamically attached to these things in interesting ways. This is super compelling. This is a market that didn't really exist before just because it wasn't actually possible to explore this way.
--:-- As well as that, we've been fascinated with how at the top level, with existing artists or existing producers, this can work as an ideation machine, a kind of well of infinite creativity that you can just pull from for ideas. Maybe you have the beginning of a track, you have a riff, you have a beat. You want to see where could this go from here. If I remix this a bit, what are variations on this? That's a super compelling thing to do as well, because it's something that before took a lot of time. It just accelerates the creative experience for professionals like that as well.
--:-- Host: I have yet to meet an artist who's actually used the products that is worried about the products competing with them. The biggest worry that I hear over and over is that somebody is going to take them away. Diego from Krea with a great reminder of just how monumental this shift is.
--:-- Diego Rodriguez: I was a creative myself doing graphic design, photography. I even tried to make video games in Flash, motion graphics in After Effects, digital sculpture in ZBrush, 3D modeling for architecture visualization. I felt the fear of, "Wait, what's the point if this thing can do everything?" But I don't think that's the case. What I think is happening is we're just giving so much power to creatives. Things that were a job in a way, now you don't even think about that. That's what technology does. One day it is a lifetime work to move from the East Coast of the U.S. to the West Coast, and people die in the process. Now you're like, "Oh, it took me like 20 minutes at the line to get to the airport." You don't even think about the fact that you flew like a Greek god through the planes. Instead, you're just thinking at a higher level. You're just flying between coasts to make bigger things. I feel like the same is going to happen. Suddenly, coloring 3D models through texture things and all these repetitive things like sketching and whatever, you will save so much time of your life. Because of not having to do that, you can focus on having even better and crazier ideas. I'm really, really, really excited to see what the creatives are going to be able to do.
--:-- Host: All right, that's all for now. The demos shared during the day were followed by a gallery party at night, showcasing many of the artists' work with the broader New York City creative community. If you want to get up close and personal with these tools, head on over to a16z.com slash AI art to check out their demos and more. We'll leave you with a little sneak peek.
--:-- [Audio clip begins]
Speaker 1: Ladies and gentlemen, I am thrilled to be here at the A16Z Artist Retreat.
Speaker 2: Yeah, it gets you pumped.
Speaker 3: So pumped.
Speaker 4: So pumped.
Speaker 5: You can generate whatever you want.
Speaker 6: That was amazing. Yes.
Speaker 7: So it's a whole body swap.
Speaker 8: Wow, this is so good.
Speaker 9: We're also working on a new type of meme.
Speaker 10: Actually, I think it's better if we see it in slow motion.
Speaker 11: Laura, you should come see this.
Speaker 12: This changes from being deterministic to being totally random. Remix.
[Audio clip ends]
--:-- Host: If you liked this episode, if you made it this far, help us grow the show. Share with a friend. Or if you're feeling really ambitious, you can leave us a review at rate this podcast dot com slash a16z. Candidly, producing a podcast can sometimes feel like you're just talking into a void. So if you did like this episode, if you liked any of our episodes, please let us know. We'll see you next time.
