00:45:15 · John Coogan, Luke Igel
[00:45:15] S0: Mhmm. [00:45:16] S3: For OpenAI, for Claude Code, for everything. Yeah. Because we think that if you use $10,000 of Claude Code tokens, which is not hard to go through, it would mean a lot if you were to spend those doing productive work in video in our platform. Yeah. That means that what can we do to help? Agents make a lot of mistakes if they are left to their own devices in this weird and odd domain, and nothing really has a URL. We had to have to be the ones to make a URL for every single timeline item and media item, and we have to be the ones to manage this long tail of problems. You cannot drag and drop the video the video files that come off of your guys' cameras into Chrome. Yeah. So we had to make it so that you can... We had to build all of that... Those custom decoders. Yeah. You know, everything from black magic to red to a few more one... A few more fun ones. But there's a long tail of tasks that go into video editing from colorization to really in... Intense performance engineering that we're here to do for you. That's why we consider ourselves managing.
00:46:14 · Jordi Hays, John Coogan
[00:46:14] S2: Yeah. Video is an an interesting domain because it's it's like not... You can't just one shot it. Right? Because you could hire the most talented, like, video team in the world. Mhmm. Like... And you can tell them, here's my objective. Mhmm. Here's my influences. This is what I want it to look like. This is what I want it to do. No matter how good they are, you're gonna have a bunch of feedback and things that you wanna change. And I feel like that's quite different than some of these other domains where you can one shot like like [00:46:43] S0: You define a goal. [00:46:44] S2: You can you can... There's a lot of different things in software that you can effectively one Yeah. Basically say like, this is the feature. This is what it needs to do. This is how it should interact with the rest of the software. And then it just does it. And it's like, okay. It functions. Whereas like, video is so... Is a... It it can... There's always... You're always gonna wanna change things or at least for a lot of different Yeah. Formats and outputs.
00:46:44 · Jordi Hays, John Coogan, Luke Igel
[00:46:44] S2: You can you can... There's a lot of different things in software that you can effectively one Yeah. Basically say like, this is the feature. This is what it needs to do. This is how it should interact with the rest of the software. And then it just does it. And it's like, okay. It functions. Whereas like, video is so... Is a... It it can... There's always... You're always gonna wanna change things or at least for a lot of different Yeah. Formats and outputs. [00:47:08] S0: Talk a little bit about... Cinema. Yeah. Cinema. Talk a little bit about, like, how close is Chrome to actual, like, native app performance? Because a lot of just rendering out a timeline can be vastly different if your GPU accelerated versus doing on the CPU. Like, there's a lot of... When you're when you're deep in the Premiere or Blackmagic rabbit hole, you actually wind up starting to think about the specs of your computer a little bit more than I think most creatives would like to. But Chrome feels like there's been a lot of progress there with WebGL and Assembly. But is it 90% of the way there? Is it a 100%? Like, where where are we? [00:47:54] S3: Well, there's this new... Just like at WebAssembly, a lot of new businesses like Figma and many others. Web codecs is what I... Is... It's what allowing this formulation of our business. [00:48:03] S0: Okay. [00:48:03] S3: It is like FFmpeg in the browser, fully GPU accelerated if you want, which means that it's here. Like, we we could not have done this business in 2024. Interesting. This has all been here pretty recently,
00:48:03 · Luke Igel, Jordi Hays, John Coogan, Jay Madheswaran
[00:48:03] S3: It is like FFmpeg in the browser, fully GPU accelerated if you want, which means that it's here. Like, we we could not have done this business in 2024. Interesting. This has all been here pretty recently, [00:48:15] S2: and Yeah. [00:48:15] S3: We're building on top of that to do the really crazy stuff where, you know, vintage camcorders that people in Bushwick love to use, that works in Kino now because we built those decoders on top of this new technology that every browser has now. [00:48:28] S0: Wait. But what what... What's different about a vintage... Is it doesn't just give you an m p four at the end of the day? [00:48:34] S3: It's... A lot of stuff is from the pre m p four era. And, of course, even kind of fancy podcast sets, they're not exporting h two six four there. [00:48:41] S0: Sure. [00:48:41] S3: You know, you have to do the ProRes and everything. But I think the the much more profound thing is that performance. If people scrub back and forth and it feels crappy, they're just not gonna use us. [00:48:48] S2: Yeah. [00:48:48] S3: Yeah. And that's what we've been grinding along... Grinding on and just pulling people in. We talked to Tyler last year when we were much earlier in our journey. His very early feedback was so useful. [00:48:58] S0: And That's awesome. [00:48:59] S3: And very grateful. Towards where [00:49:01] S5: we come.
00:49:01 · Jay Madheswaran, John Coogan, Luke Igel
[00:49:01] S5: we come. [00:49:01] S0: Yeah. That's amazing. How are you thinking about integrating AI video generation, more generative AI tooling? It feels like we're oddly still in the slop era. There's some projects that I'm seeing coming on the pipe where it's like, okay. This might be the time when we all sit down and collectively watch an AI generated film and think it's good, but we're still not there yet. Like, Harry Potter Balenciaga was, like, three or four years ago, and then there's there's an AI movie, The Gods Must Be Challenged or something. I forget what the name is. But there's there's an AI movie that's coming that people are optimistic about, but we're not quite there. But it feels like a hybrid solution is fully embraced where it's green screen replacement, background, set extension, that type of stuff. Are you thinking about integrating those tools seamlessly? [00:49:55] S3: Like like so much knowledge work, I think LLMs are are gonna be just so important here. Seeing how Astra is such a good collaborator in things like Blender [00:50:03] S0: Yeah.
00:50:03 · John Coogan, Luke Igel, Jordi Hays
[00:50:03] S0: Yeah. [00:50:03] S3: Is exactly what I experienced alongside people the I made this video with where we just did crazy amounts of visual effects that I never knew how to do before this. Right? And if I could choose between some magical lab who could give me a model... A diffusion model that could make a Pixar like movie or a, I would say, coding agent that could use RenderMan and Blender and video editors to make a Pixar like movie, I would choose the LLM every single time. Mhmm. There's a level of controllability that I think all real artists clearly wants. I think the method... The model of AI movies as you type into a text prompt, and then you get a beautiful movie at the end, just kinda feels like the apocalypse for a lot of people and for good reason. And there's a there's a kind of positive vision that we're really excited to demonstrate here where it's like, no. Everything is code. Everything is, But at the same time, everything is what you remember. The... This part we're seeing right here. Yeah. We just had Kino reskin all of Kino itself. [00:50:57] S0: Sure. [00:50:57] S3: We downloaded the ISO of macOS Leopard and then reskinned Kino to look like that. Wow. And then my friend interrupts me and asks me if I'm gonna stop goofing around and ship the video and [00:51:07] S0: Yeah. [00:51:08] S3: All that. [00:51:08] S0: So how much of that was made in Kino? [00:51:11] S3: All... That was all Kino. [00:51:13] S2: Okay. Kino.