Welcome back to the Will Ventures podcast. My name is Kushana Huja. Today we're speaking with Cristóbal Valenzuela, co-founder and CEO of Runway, a company helping define what storytelling looks like in the age of AI. From pioneering video models to building tools used by filmmakers, studios, and creators, Chris and his team are pushing the boundaries for how to convert an idea into a visual experience. Today you'll hear about the rise of world models, the future of creativity, and how AI is impacting sports, media, and entertainment. Enjoy.
Well, Chris, thanks so much for joining us today. Really appreciate you making some time to join us on the Will Ventures podcast. You know, to get started, it'd be awesome to go through your background, your journey, where you're from, how you ultimately became an entrepreneur and started building Runway.
Yeah, of course. It's a long story. I'll try to make it short and condensed, but the short story of Runway is I'm from Chile. I grew up in Chile. I studied there. I've basically spent most of my life there. And then when I was like 26 or 27, I got a scholarship to study at NYU. And my bet at the time was, this is like 2016, so in AI years, it's probably like the 1700s. My thesis at the time was, look, I'm seeing some interesting stuff with neural networks, deep learning seems to be taking off, and I want to just first of all understand it better, understand what it means, but also understand the implications that it has on creatives and on art, on media and film. I was studying business and econ, and I was dabbling in art on the side. I was learning and teaching myself how to code. There was a bunch of different interests at the time, and I was like, well, the only way I could really expand on all of this is just spend at least a year or two studying and focusing on this. And so that's what I did from 2016 to 2018. I had the fortune of meeting my co-founders in school back then, and it immediately became obvious that we're all obsessed around the same ideas. It's like deep learning is starting to pick up. We're very much into understanding what you can do with these models, how you can creatively think about using them. And again, this is for context for everyone who hasn't been paying attention over the last 12 to 15 years. Speaking about AI images or videos or AI being able to do the stuff that we know it can do today sounded insane. The state-of-the-art in 2014, 2016 was image recognition models that can distinguish a hot dog from not a hot dog. But we were obsessed with this idea that if it works, if you can generate the stuff that now we can generate, it just felt like something we should try. And so that's what we did for a lot of time at school. We just spent time understanding, researching, and then after we graduated, it was clear to all of us there was one thing we could do, which is start a company and spend all the time we could just trying to solve this problem.
Did you have a background? Did you, you know, becoming an entrepreneur, taking that step, starting your own company versus working somewhere else, like what was the decision point where it was like, no, I can't do this elsewhere, go join a lab or research entity, like we need to do this together because this problem is so unique or so different, etc.?
Yeah, I think we had chances of working at larger companies or we could devote some time to think about these problems and how we could perhaps develop them with some offers to join some companies to do it. But I think at the time we were like, this is just too interesting to do it somewhere else. We like working between us, we like the kind of projects that we build. And I didn't have entrepreneurial experience. I've never raised money. I need you to understand I was an F1 student two years out of Santiago in Chile, and that felt daunting to be honest. It definitely feels like there was a lot of uncertainty about, well, how do you do this? I've never met an investor in my life. There's many things that now are obvious but back then they weren't for us. And I remember at the time a lot of people were telling us, well, what you should do is you should just get a job at Google or whatever and spend six years there, get a green card, and then build your network, build your relationships, start to meet people in Silicon Valley, in San Francisco, meet investors, raise some capital, and then maybe you'll figure out the business. It's like, I don't have six years. I have like six days, six weeks. I don't, it's too much. Maybe it's my impatience and my anxiety. I just know I want to do it now. So we crafted this internal way of operating, which was like, well, let's just figure this out. How hard could it be to do this? And I think when you're naive enough to understand that it's way harder than you think, you don't understand how hard it could be, you're much more willing to do it. If I knew how hard it was, I might have doubted it, but now it's like there was no other choice at the time. We just went with it and we learned on the go and it's been very rewarding after almost seven years now.
So take us to the early days seven years ago until today. Maybe talk a little bit about the product evolution. And I think one thing that's really interesting as we've gotten to know each other is, you know, folks know Runway because of the leading research you guys have done in AI video and the models that you guys have built that are proprietary, but there's also a lot of software and workflow tools that you guys have created for the creative industries. So, just kind of walk through the business from the early days in terms of what you guys are developing to what it kind of looks like today in terms of the different products and things you guys have for your customers.
Yeah. So, that's a good question and I think a lot of it has to do with what are you optimizing for. I would say back then at the very beginning it was still unclear how you solve for the things that now you could solve for, which is large video production or large content or large video models in the way they work now or world models in the way they're working now. And so we did a lot of research and exploration back then. We still do a lot of it, but back then it was much more experimental than I think it is these days because it's not only research, so science. You're trying to create something that's never been done before and you're very methodical about it, but it's also a new market. Like what's the market? I remember some of the best investors I've ever met literally telling me there's no market for generative AI. Like the word you just came up with, generative AI, or I was calling it synthetic media or synthetic AI. It's like a made-up term, where does this fit? I build a mental taxonomy and a mental model and a map of companies, this is again 2018, like here's the design companies, here's the video companies, this thing doesn't fit in any category. So, you know, what's the time of it? But that for me is the point. That's exactly the point. This is net new and it's less about incremental work and it's more about, look, if it works, you're creating entire new things. And to do so you have to first of all work really hard but also experiment and try and explore. I think we like this idea of exploration a lot. And exploration from a research perspective is, from all the choices and decisions you can make, which are you going to pick and for what reasons. And there's a lot of conversation these days around taste. I don't think taste for me is around aesthetics and how things look and how you see the world. It's more about when you need to make decisions. Which decisions are you going to make that are probably unpredictable and obvious but highly relevant maybe a few years in. And I think we've fine-tuned or we've built a better intuition of taste over time, research-wise and product-wise. And so we do that and as we do that, I think we've also learned over time how to abstract and build products around those new exploration things. So again, we're the first ever to create production commercial-ready video models and we're the first ever to create a UI around it. And now the UI becomes obvious to many. It wasn't obvious to many before. And I think that's just, you need to understand the insight, the customer, the need, the goal you're trying to solve for. And this is something, if I look back, I think the company has been in a highly privileged position because we've been basically always trying to merge both art and science and we're good at the combination of those two worlds. And being good at art and science requires you to be in some ways a polymath. You need to know about art, you need to know about the history of art, you need to know about creativity, art, content creation, film, many of the things you're touching. But also science. Science operates differently. There's a scientific method you have to know and care about. And I think maybe a good summary of it, we've been good at transitioning between the two of those.
Yeah. That's amazing. So maybe you spoke a little bit about the customers there. You know, one of the things that I think we've really admired about you is you're one of the few folks that have been able to build real relationships with the entertainment community and build that over time. And that's not an overnight thing, that takes time, that takes explanation and going there and meeting folks. But you've also got this really interesting set of customers that are more aspiring filmmakers, consumers, prosumers, etc. Just kind of walk through like what is the, you know, who does Runway serve today? Both on maybe like the enterprise and larger scale side as well as some of the folks who are just starting to get into kind of content creation, filmmaking, etc.
Sure. So we have a wide spectrum of customers and use cases and users that range from every Hollywood studio that you can imagine to advertisers, brands, marketing teams, creators, consumers, social media teams, and just individuals. On the enterprise side, we have a full offering and set of products and models that are specifically tailored towards people who are making content for a living, people who make either ads or professional content for films, for narrative work, for just short-form content. And the goal there is how do we make you get what you need faster and better than ever before. So if you want to make an ad and the ad involves having a studio and having actors and getting audio design, well, you can just one-shot that thing. Now it won't one-shot the entirety of the ad because you need a concept, you need the art direction, you need the idea, but once you have the idea, you can go from idea to final completion of a video or a concept or a final ad in just a few minutes. And that becomes incredibly useful as you can imagine for the best filmmakers in the world but also for advertisers and other types of format and content to just people who want to post fun stuff. So we do a lot of that and then there's other avenues and other markets of the business that we're expanding into. And this is more just how models are operating these days where these models at a general level are basically generalization machines. They can understand a domain but then extrapolate the knowledge of that domain into many others. And so video is for us a simulation problem. You want to simulate how advertising works, you want to simulate how filmmaking works, how storytelling works, and you can simulate them pretty well. That's the conversation of like will this ever work around storytelling maybe was there six years ago. I think now it's pretty obvious that it works. But then now you can take those models and start thinking about them in the context of robotics or games or AV where video becomes, pixels on a screen become incredibly useful as well. And so we're also kind of opening that part of the business.
100%. We'll get to the world models and that point in a little bit. You guys work with folks like Adobe, with Lionsgate, with Netflix reportedly, Shutterstock. I mean just a ton of really amazing folks on top of what you guys are doing, you know, with Nvidia. I'm just curious like with these customers and when you were initially kind of talking to them, like when did you really feel like you guys found some product market fit with those customers? What was really resonating in the early days and even today like when your team goes and speaks to some of the folks on the enterprise side in terms of where do they kind of see some of the magic in what Runway has built?
Yeah. I mean it depends, all of those sometimes use cases are different. I think for some of them it clicked earlier than for others to be honest. I think for studios it took a bit more time because models were not fully there yet when we started chatting with many of them. I think that's one of the challenges with AI these days is there's too many things happening but a couple things that I've seen become kind of obvious these days is that model capabilities are advancing faster than people can catch up to. And so diffusion of technology is just happening extremely fast and models are advancing extremely fast. But there's still sometimes a gap between what enterprises know and what they want and what they think the technology is able to perform and do versus what it actually does. And a common trap I've seen many fall into is you start, okay, this conversation has happened in many boardrooms in many parts of America, in many parts of the world actually, like what's our AI strategy, how are we going to start using models. And so some AI department is spun off or some leaders are set up to solve a particular problem to start using AI in some way and then you find a company you want to work with, a provider or a model. You scope something, you work on it for like four months, you get procurement, this is again we're talking about huge companies. So you get procurement processes approved, you get security reviews, the whole thing. You're nine months in, you're like, okay, we're finally ready to start using Gen 2 and Gen 2 is deprecated, we moved on, the stuff that you can do now is very different and much better and also works differently. And the company wasn't prepared for that level of speed and they thought they were moving fast but really they were not. And that in itself presents a problem because now you need to either start that whole new thing again or just wake up to, look, if you don't change internally the processes, you're going to be left behind. So for a lot of our customers, that realization, that conversation and that process happened and many of them realized, look, the models will do the things that they do right now. We need to just move faster and start adopting them and figuring out how it fits within your arc because the content that Netflix does and the way they work is different from the way Disney works and it's different from the way Lionsgate works. There's some things that overlap, some of the same companies and providers they use, but overall they work differently. And so they all have very specific different ways of using the models. And that's for us a big differentiation is we work with them to help them identify how it works. So to your answer, I think it depends. It depends on where the company is in that journey and where exactly do they want to go.
If you're a creator today, if you're a talent, if you're an IP or rights holder, this industry, as you mentioned, is changing really quickly, like Gen 4 to Gen 4.5 and whatever's coming next now with Runway as an example. I'm curious like how should those folks be thinking about AI video, using it in their day-to-day? How should they be thinking about adopting certain models versus others? Like it's a rapidly evolving space. So, and you're kind of in the thick of it. So I'm curious like for those types of users, the creatives, the marketers, the creators themselves, the talent, like how do you kind of think about that?
I think you need to have solid principles that guide how you just navigate the technology and the landscape. If you're too obsessed around model capabilities and what the models can and cannot do, you're kind of missing the bigger picture. And the truth is that the models are really good at a lot of things or not that good at some other stuff, but overall they're going to get better at everything. That's my principle. Assume that most of the things that you're doing by hand across industries, across economies, across segments and markets are probably going to get either fully automated or fully accelerated with AI. And so if you're at a point where you try Gen 4.5, which is the best model right now, and you created this amazing scene and you look at the scene and there's, well, Chris, there's a character inconsistency in the middle of second four and therefore I can't use this, and be like, yes, maybe that's true but that's going to get fixed in a couple weeks, months at the most. And so either you start understanding how this fits within what you do or by the time you do so, you're already behind. And that's been the common thread in AI video, image, and just media generation for a long time. I mean we sometimes argue that we have short-term amnesia as a society because the biggest discussion, and this is true, there were op-eds written in the New Yorker and many other places that people had a year and a half ago was that AI was not able to render a hand with five fingers. It was like a funny thing, it was like a provocation, like you see, it's not going to work, it's slow because it can only make hands with six fingers. Unless you're living under a rock over the last two months, I think everyone who was arguing that is very quiet these days. So take that and think of the maximum version of that, 10x that, 100x that, and you realize that many of the things that you can do today will be generated, even the things that sound insane, like long-form narrative content that's coherent, consistent, really good, structurally from a story perspective, will get one-shotted generated.
So talk us a little bit about your product philosophy and kind of the framework as you guys think about in terms of building new products and building net new kind of things on top of Runway. So you guys have a deep research base and you've kind of always relied on that to grow the foundation models to ultimately continue advancing the technology. You've built some of these interesting kind of creative workflows. Like what's next for Runway as we think about net new products and things that you're going to be kind of launching to your customers and what kind of gets you most excited about the product roadmap ahead?
Oh, there's so many things. So on the philosophy, and I could answer this without going into too specifics for obvious reasons, but kind of what I was touching before, I think a common framework that we use for product and we tell the team around a lot is you can think about any decision that you want to make, and this actually stems from AI research, between a decision between exploration and exploitation. So think about anything, like you want to get a coffee. You can choose two kind of strategies. One strategy is you're going to exploit and optimize for what you know, which is I know the coffee shop right across the street. I'll get what I always get and it will be quick, fast, and cheap. And exploration is, well, I heard about a new coffee shop that's two blocks away. It's harder, it's longer, but maybe I really find something that I never knew I liked or I meet someone I never knew or I discover something as I walk through. It's like you're opening yourself to newness. And you can think about any company out there and it roughly fits either one of the two. They're either very optimized-driven, very economies of scale, or they're innovation-driven, creating new things. Today in AI, and if you want to be a frontier research lab and a product company that's relevant, you have to be ambidextrous. You have to be able to use both hands. You have to be able to navigate long horizons and long strategies that are years in. That's the exploration part. Those bets are culturally different from the exploitative part, which is how do you monetize, win, and continue building, attracting customers for a product that works today. Now, if you just focus on one or the other, you can also think about this in many AI companies these days that some of them are just focusing on whatever works today, we're going to build a product around it and monetize it. And I think that works for a very short time until exploration basically leapfrogs you. We're seeing this with code these days where maybe a great bet, an excitingly really well-timed bet was to bet on IDEs a year ago because that was the right function format. But if you only betted on IDEs and you didn't explore enough, you were going to get leapfrogged, and that's exactly what's happening right now with cloud code and Codex where I code on just language, I don't open my IDE anymore, I just don't need it. And so if you spend too much time on exploitation, you're going to be left out. Now, if you spend too much time on exploring, you don't have a sense of reality. You're kind of checked out and you're doing something that it's either an extremely hard bet that it's either going to work or not. And I think today within how things are progressing, I'm not a fan of that. I think you lose touch with reality in a way. And so our product philosophy is, well, you have to be ambidextrous. The culture of the company has to be set in such a way that would allow you to do both things. We have research teams that are pushing frontier science that's like many times what I could even start to understand myself, which is amazing. And you have very specific teams that are much more optimizing for what works now and growing what works now. And those two things need to be culturally aligned.
It's awesome. We have a long legacy and history in sports at Will Ventures. Sports historically has been a tough end market to crack for emerging and net new technologies. I'm curious if you have a point of view on your business and just AI and video broadly and how it might impact the sports industry going forward given so much of the value is in live and media rights etc. But curious if you guys have kind of explored that end market and have a point of view on where the business is going as it relates to sports and AI.
Yeah. I mean there's a lot specifically when it comes to nonlinear or real-time video generation or media content. A lot of people's experiences with AI video these days has been async. So you type something, you create something, it takes two minutes, you get a video back, fun or great or edit, whatever. You need to understand that where things are heading is you're going to be able to do that in an almost real-time or real-time basis. So, I can not only prompt for a video, I can have a conversation with the video itself. And that could literally be in the form of I'm going to chat with an avatar, an agent, a character that I'm having a natural conversation with. And you will totally forget that who you're chatting with is not real, which is kind of amazing. But think about sports, right? So entertainment or sports or any live event where you're not only consuming what's happening but you can also have a dialogue with it. You can chat with a player who scored a particular thing and you can ask him about the history of other things he's accomplished. You can also have animated characters be part of it which I think for newer generations becomes really an interesting avenue of exploration. And then what you can also do is there's ways of augmenting the content that you already are seeing on a screen. So moving the camera in ways that are newer or different, generating new points of views and just overall making the experience a bit different, more engaging is something that I will definitely, is already happening within sports, but I will definitely start seeing accelerating over the next couple of months.
Maybe taking a step back, let's talk about the market broadly. I feel like in the last couple weeks there's been a ton of viral videos that have come from Sora and others that have been launching net new products and consumers have been playing with it. I'm curious if you have a take on what's happening there with Sora and also just the competitive landscape around companies that are both building deep research foundation models on the video side but also have a consumer application that's got a social interface, a la Luma, but just how you kind of think about Runway's positioning as it relates to the broader landscape.
Yeah, I mean I tweeted about it like two years ago, I was like, look, video models are going to become the most interesting and discussed technology in two years and I was off maybe by a few months. What I mean by that is it's becoming a national security or a national public cultural interest, and this is way beyond us Runway building video models and working with studios and IP holders and everything else. I mean now you have nation states working on this and it works pretty well. I mean those models are becoming really good and I think it's expected. If you open a market and you open an industry and people are seeing the value of it, you should expect the biggest players to come on and play with it. And I think what we're seeing is that there's different values on how you build this technology. Maybe this is the best way to think about it. We have a specific set of values around humans and human creativity and how do you augment a creative process with AI models in ways that are novel, new, and different. Some other people may have other values around how you build it and you can see that being shaped and formed in real time and you can see models in a way, I think models for me are representations of the organizations that build them. Like language models, you can tell they have a personality. Using Gemini is different from using Claude and different from using ChatGPT because I think they were built by different organizations with different sets of values and those values are visible in the personalities of the models themselves. That same analogy or way of thinking extrapolates and applies to video models. The way you build a video model, the way you make it accessible, the way you prompt it, the aesthetics, the art direction, the kind of things that work well and not well are a reflection of the organization that built it. And I think this is now become visible in Sora and many other of the models out there that they just operate on a different value spectrum that we do. I think we might need to choose which value systems we want to play with and have as a center of how we develop this.
