00:02:26 · Anastasis Germanidis
[00:02:26] S2: Yeah. So so right out of college, so I studied computer science. I was, you know, I was... I've been programming since I was 10. So it felt like the a natural thing for me to just go into San Francisco and in the tech world right after graduating, but I did find like there was a level of like a monoculture and a way of building things that were like I really always like to turn my ideas into projects and like had like always you know growing up like 20 to 30 different projects going on at any given point and what I saw in San Francisco when I lived there right after graduating was there was kind of this intense kind of drive to like start a company and commercialize everything you're working on and I really for a long time I really you know what I cared the most about was just this idea of like really turning my ideas into projects and really caring about the craft of like software and the craft of creating things and so decided to and and I also like technology as as a way of like building technology as a way of questioning and as a way of really thinking through like, you know, how does these things that we're building like kind of change society? How how how how how does technology kind of can create like... How how does creating things basically help empower people, give people tools to allow them to do something they couldn't do before. So it felt like there was a big pressure to create a company around everything that you're creating, But I I felt I I would need more time to explore basically. And so I decided I wanted to, you know, become an artist, go to New York, go to art school, figure out, kind of explore and figure out like what I wanna really focus my life more on versus trying to be in the tech world from the get go.:
00:04:37 · Adam Miller, Anastasis Germanidis
[00:04:37] S0: And did you have deep artistic influences growing up? Were you particularly drawn to say film or literature or painting or something? [00:04:46] S2: Yeah, too many. So I was I mean as an adolescent I was spending kind of all my time a mix of coding, writing poetry and making music. I didn't have a good way of combining all those things. I started trying to combine them in college actually. Made one of my first projects was using a technique at that point that was very popular in ML and AI called genetic algorithms to figure out how do you make a system that can draw pictures of faces. But it was extremely early at the time, so it took quite a bit of time, I think, before image generation got there. But yeah, was always very interested in this idea of like systems that can create art, systems that can draw, that can kind of help with a creative expression. [00:05:37] S0: So the founders of Runway don't really have the storied Silicon Valley foundation model typical backgrounds. You've instead described yourselves as three failed filmmakers. How did you all come together to actually build this company?
00:05:37 · Adam Miller, Anastasis Germanidis
[00:05:37] S0: So the founders of Runway don't really have the storied Silicon Valley foundation model typical backgrounds. You've instead described yourselves as three failed filmmakers. How did you all come together to actually build this company? [00:05:51] S2: Yeah. So it's it's interesting because what we found out, and that wasn't something we discussed originally, but like all three of us had some kind of initial contact with the world of film and initial attempt to enter it. I originally, when I went to college, I wanted to study filmmaking and neuroscience. Those were the... I wanted to get away from coding and computer science and just focus more on kind of those two areas. And I quickly found out that the... Well, I like the idea of making films and I, you know, I I wrote screenplays. I I was kind of trying to kind of work in that world. The It was less of like the, you know, being on set and being a director and like really creating films was not like the thing that I would enjoy the most. And I think that's, you know, there is similar stories with my co founders of either going to study film or making... Having side jobs of making videos. So we always were very inspired by the world of filmmaking, very inspired by film. I did a bunch of, like, collaborations with filmmakers where I saw the way they were working and kind of wanted to figure out how those... How to use technology to help them. But, yeah, we ended up arriving at the world from from a different direction.
00:07:25 · Sachin
[00:07:25] S1: And take us to the specific moment when you realized that you could build video generation models. Because I think something interesting in studying the background of the company is that you guys were actually quite early to this. I would have assumed that you started in 2022 after kind of the chat GPT moment, but you guys started way before that. Can you talk us through how that moment came about?
00:07:46 · Anastasis Germanidis
[00:07:46] S2: Yeah. So when we started, generative AI wasn't a term. Like, we were trying to figure out what the term... The right term was to describe what we're doing. We were coming up with, you know, synthetic media and, like, trying to figure out how to describe this thing. So we started out... Really we started collaborating with each other around 2016 when we all were in the same kind of art school. And we're initially just making... You know, at that point there were some early very interesting demos of applying deep learning in creative ways. But it was very, very difficult to use those models for if you were an artist and a filmmaker. And so the initial goal was let's let's make those models very easy to use in an interface that feels like a creative tool, but it's actually powered by those new kind of early image and video generation models. We quickly realized that in order to advance those models, in order to give them the control that's needed, we needed to build a research team in house and build a research organization but yeah it took quite a long time before you know we got anywhere close to results that were kind of usable in a production setting It took maybe from like, you know, twenty twenty eighteen when we started to 2022 or 2023. So it was it was kind of a long process of really having conviction that video generation, we we get there. And a lot of it was just driven by just seeing the improvement of those models every year. So you had like... You started from, you know, the early... What was called the generative adversarial networks was one of the earliest kind of neural network architecture for image and video generation. And you could see... You started from, like, 32 by 32 pixel images. And every year, you have, like, resolution, you know, double fidelity and quality improve predictably. And so you could see that if you extrapolate from that, we'd get to the point where you can make really compelling, like, realistic outputs, but it took it took it took a while to get there.
00:09:54 · Adam Miller, Anastasis Germanidis
[00:09:54] S0: Mhmm. And fast forward from when you started in 2018 to where we are now, eight years later, Runway isn't just being used by individual creators. You've got film studios, ad agencies, indie studios using these tools. How would you describe what people are actually using Runway for nowadays? [00:10:12] S2: Yeah. Runway has been useful for all the things that you mentioned and many more. So we think of the tool as being a very general purpose tool. Right? You can create images and videos for any kind of purpose. A big set of use cases in... Are in media entertainment. So we have all the major film studios using Runway as part of basically every part of the workflow. So you have, like, from preproduction to production to postproduction. So you might use Runway for concepting and storyboarding for essentially kind of blocking out, like, what the... Your short or your ad looks like to... Increasingly, we're seeing runway being used for the final frame as well. So the thing that you generate in runway actually being used for, you know, the the thing that you distribute and share it with the world. And then a lot of use cases in post production. So in, like, VFX work and editing and background removal and, like, all those things that are very time consuming but very... Part of every... Basically, every VFX workflow. So, yeah, Runway is being used in in film. It's being used in, in marketing and and advertising. It's being used in, in games, in architecture. A big part of our thesis is that video models are very general, and they can be used in all those industries.