00:28:26 · Amit Jain
[00:28:26] S0: Yeah. So actually, I'll tell you a couple of things. We were at this gaming company, Savvy Games, that makes, like, you know, Monopoly Monopoly Go and and huge number of mobile games that you you you guys might actually love and and play a lot. So we were we were talking to them and, like, you know, they were talking us... Talking to us about, like, know, hey, they're doing this thing with Avengers where like, know, they will have characters from Avengers in their games. Yeah. And then creating every single one of these takes like, you know, months and months of work. And I was just sitting there, like, know, they were talking about they were talking and then I I downloaded some of those images from from the Internet and like, you know, I asked it like, hey, look up every single Marvel character and, like, you know, produce a campaign that can be done in, like, you know, 16 countries, like, you know, with different languages. And by the time we had, like, you know, in about thirty minutes or so, I was not really paying attention to the board or that is the product we have. It had produced something really incredible actually, and we were just like, oh, let me plug this in, and we were showing it to them. And it was actually ridiculous for them to even imagine because they were like, oh, yeah, we thought that AI could be fast or all these kind of things, but when they actually... When you actually see it in practice, that is actually ridiculous. The second one is this Ben Kingsley production now. So I was with James Cameron and John Favreau this thing in LA. Essentially the Academy of Arts. Anyway, so this was a long winded conversation about AI in Hollywood and AI in movie production and studios and all these kinds of things. And oh, where can we use it? We can use it for like in this tool, this tool, this tool. But everyone has a very theoretical idea. So when we started to show like, okay, look, this is how they act. And in about like five minutes, there's a full production loop, a thing that would have taken six months before, A full production loop from like shooting to editing to changing the whole set and then being able to produce the thing. That blows people's minds. And again, it's not really easy to blow James Cameron's mind because he's the director of blowing everyone's minds really.
00:30:25 · Amit Jain, Finn Meeks
So I think that was actually a really powerful moment for me that like even people who you would imagine like are are pretty blown can be actually... You know, you can you can grow their minds quite a lot. So I think these are the kind of magic moments. [00:30:38] S1: That's that's amazing. Those are some great stories. I... I'm curious, you know, obviously, you guys are are blowing people's minds kind of in, you know, large enterprises. You've had a lot of, you know, individual creators using the products in the in the past too. You know, there was obviously a pretty big wind down of a consumer video kind of visual kind of effort recently with Sora. I'm I'm curious, like, where do you see these models kind of being most useful? Is it enterprise? Is it individual creators? Is it consumer again in the future? I'm just curious how you how you think about that as a founder.
00:31:16 · Amit Jain
[00:31:16] S0: I mean, we have had efficient video models for a very long time and like, you know, we talked in our teams many many many times about building like a generative social network or something like that. But the problem we always would come back to and like, you know, if you guys know our teammates, you should ask them. We talked about this in 'twenty five, and it's '24, 'twenty five, all of those times. But the problem was what makes a video interesting has almost nothing to do with whether it was generated or not. So when it launched, most of you probably had the same moment where it was like, one, it was really interesting immediately. And you were like, oh, this is generated video and then generated video can be generated. You scrolled for maybe a day. If you're really dedicated like some people who are in our teams, they scroll for two days, three days, five days, But then it's like, okay, now what? Right? And so you saw that in sort of usage graphs and retention graphs, all these kind of things, because the problem is that a video is not interesting because it's stun rated. A video is not interesting because it's CGI. That there's almost nothing to do with why it is interesting. Some of the most interesting videos on the internet are the lowest resolution, right? Like the worst kind, but they are interesting because what's happening in them. Until we solve intelligence in video, a network like Sora is pointless because you need these models to understand humor, you need these models to understand context, you need these models to understand your context. A video that might be interesting to you, I might find like, what is this? I have no idea what you're looking at, right? Because context is local, right? Humor is local. Something we might find funny in San Francisco, people in Miami like, what the hell are you talking about? Right? Like, I can make a joke about MOE and like, you know, you guys might laugh. And in in Miami, I was like, this is a weirdo.
00:33:01 · Finn Meeks
[00:33:01] S1: They made an entire TV show about it.
00:33:03 · Amit Jain
[00:33:03] S0: Exactly. And I hear there's a new one coming out. It's not called Silicon Valley, I hear there's a new one. Anyway, it doesn't matter, I forget the name now. So it's really context, and that is all intelligence, really. Being able to tell a joke that other person might find funny requires understanding the state of that person, Requires understanding the environment they're in, the local context, all of these kind of things. So we didn't launch a sort of like thing primarily for that reason because we just didn't find it interesting, right? The videos that Rip was generating and I think that's what the world founded too. So consumer, by definition, consumers don't generate or create, consumers consume, creators create. So until we get to the point where the creator is the computer and the models are intelligent enough, like unified models we're building, you're not at a point where consumers will wanna really create out of the necessity of their heart. My mom doesn't really create images. She might make a meme here or there, but she's not gonna pay for it. On the other hand, enterprises, and by the way, the businesses enterprises, they are responsible for 99% of the pixels on our screens every single day. By that, I don't mean just like the largest companies in the world. A TikTok account that is doing well, 100,000 followers, ends up becoming a business. Very quickly, they're gonna have an agent, they're gonna have an editor and they're gonna have real revenue really, actually, you can generate a lot of revenue this way. So they're gonna end up becoming a business and they're gonna have a proper operation. They're gonna go for the best tools they can find better than the guy next door. So businesses are currently our biggest customers, and I think that is very obvious reasons for that.
00:34:43 · Finn Meeks
[00:34:43] S1: Coming back to what you said earlier is you you kinda do these deployments. You learn from what your customers are doing, you feed that kind of back into research. And it goes back to something we were talking about a few weeks ago that stuck with me which is that you just said product and research are becoming inherently more coupled. I'm curious what that looks like at Luma day to day, you know, as you kind of feed this back into your product learnings, kind of back into research. How you do that maybe just in practice day to day and then what that kind of enables? I think one thing that has really impressed me about Luma is that you you actually ship and you ship with a very high velocity. And I'm just curious, can you talk us a little bit more about just kind of those learning feedback loops and kind of how you see these being coupled?