The Neo Show: Founder Fridays #15 - Ondrej Dusek with Hosts Howie Xu & Axel Pilette

This is the Neo Show where the minds shaping tomorrow speak today. Brought to you by Neo, the browser redefining what AI can do for you.

First of all, thank you for having me. I think we will have a very exciting conversation. I'm very excited about this piece. So about myself, Andre Dusk. I'm the founder of Creo AI and we are an AI platform for video making. So we built what we believe is the simplest and fastest way to create long-form videos for social media. I can talk about this a little bit later. I know that discussion is probably focused on other topics but very happy to talk about this. Previously I was a partner with McKinsey here in the San Francisco office. I was part of the digital McKinsey group. So most of my work was connected to technology and new business building for mostly US clients here.

Yeah, so I'm honestly very excited about the browser. I started using Neo like it was late last year. I believe it was late October or like in November when the first beta release came out. And since then it's been on my computer and pretty much been the primary browser for most days.

Well, I mean I love the browser. So when it came out as a beta, I mean, it was still a little bit barebone, but what caught my attention right away was I think it's called the magic box. Essentially, the integration of the URL box with AI chat. I found it very productive, very creative. I think you guys were probably the first ones who came up with this concept. And sort of that really clicked with me. I think it's where obviously now you see a lot of other browsers going that direction. So I think it was something that you guys probably brought first to the market and I think it's a great invention that caught my attention at the beginning.

That's where the future is.

Yeah. So for me it's the convenience, simplicity but mostly productivity. So I obviously I use it for a blend of my personal reading and personal search but also for my work for Creo. So it is way more efficient to search for information and digest information. So being an AI startup there's a huge amount of research and reading and kind of following the latest trends and innovations that I personally have to do and the magic box just makes it much easier.

Yeah, I mean it plus more conveniently, right? It's just like you don't have to think like, oh, am I searching for this or am I chatting about this? For me it sort of flows together and from left to right each time because let's say I'm doing a research on the latest innovations in the AI video models, right? So it always comes to a combination of searching for the latest articles and then reading longer articles and then being able to summarize and make sense out of things. So it's always the convenience and productivity that is helpful for me.

Well, absolutely. I'm using the browser. And I suspect you guys are just getting started, right? There's probably more coming out. But it's a refreshing experience to have some innovation in the browser, right? Because browser has not changed for 10 years, for so many years and honestly even when I started initially when I first downloaded Neo I was like okay browser okay let me give it a try but I wasn't expecting much excitement I was going to be another browser, right? I used other browsers before. But there's something that clicks. It's about the productivity and the simplicity of using the browser. So it helps. I think there's obviously more to come and the industry obviously is now woken up a little bit to it.

Yeah. Absolutely. So let me maybe step back a little bit why we started Creo and what we're trying to solve.

So it really started, it was a little bit personal because I was trying for many many years before I was trying to be a video creator myself but it never really worked and I have to say embarrassingly that it was a fail. And the reason why it didn't work for me is that I just didn't have the time to put the hours into it every week. I didn't have the skills, I didn't have the tools, I didn't know how to use all the video production management and editing tools. And I didn't even want to be on camera to be completely honest. And then so it comes a little bit with my personal experience and with sort of even solving my own personal problem that there are a lot of people out there in the world, for example the two of you that have a lot to say to people out there, expertise, experience, teaching, that don't have the time or the skills or the resources to do high quality videos, right? And that's sort of one part that we realized or I realized when starting Creo. The other part is that when you look at what is truly engaging at the content that is out there then you realize like these days pretty much anything that is just written has very low engagement rates. There's just so much to read people don't read that much anymore. There's two categories that have high engagement rates. One is videos and the other one is user generated content of people going to Reddit and chatting with actual authentic content. And the sweet spot is obviously the intersection of user generated videos. So that's the second part of things and the third part of things is when you actually look at how the video creation process for most people works. I mean even today it's just very laborious. It's very difficult and on average people spend an hour to edit or craft one minute of good video, right?

So in general both. So in general I think when people want to create content I think most people don't actually even differentiate like is it a video or they just want to get their idea or their expertise out there and share with as many people in as simple and as fast way as possible. And then what we realized is that previously people were spending or still are spending one hour per one minute of good output video but it doesn't have to be that way. When you actually do the homework and go through the process a lot of that can be automated and a lot of that can be substituted by AI. And then with the advancement that we are seeing in the kind of AI video quality I think it's just a matter of time before most video will actually be AI video because it's just so much better simpler and faster. And so we saw an opportunity to help people like those part-timers or those aspiring video creators, people who are starting or people who are just at the early stage of scaling their video content that they don't really have good alternative options how to create quality long-form educational or expert videos for social media. And that's why we founded Creo and so essentially our aspiration is to bring the fastest and simplest way how to get the job done.

We're certainly not trying to compete with...

So both. We also have a model and we're also using third party models. But the way I look at the market is sort of like at the bottom when you kind of visually look at the stack, at the bottom there are a bunch of foundational models, right? And V3 is there together with other companies that essentially can afford to put tremendous resources into actually creating those foundational models. Then on top of that are sort of like vertical sectors. There's a bunch of B2B players like Synthesia that is doing more like B2B videos. There's B2C players like consumer grade videos like Creo and others that then focus on different segments and other aspects of the video. And what we're focusing on and I think most of the players above the foundational models they're focusing on the workflow, right? That's the simplicity, that's the user interaction and solving a real problem and bringing real value to the user.

So where we adopted and kind of co-created and fine-tuned our own model at the end, and I call it brutally optimized, is that we needed a lower cost solution for a lot of the consumer videos. So if you for example look at V3 it's amazing the quality is just unparalleled and one if not the best on market, but the cost is staggering, right?

It's probably one of the fastest growing revenue products within Google in history.

But if you look at a regular content creator and to create a 10-minute educational video where you have to redo a lot of these video clips, I mean, you cannot afford to pay like $300 just to get a YouTube video done, right? So we were looking for finding a lower cost capability that is good enough at this point.

So the way we're... so that's obviously not the end user kind of cost but our infrastructure and GPU cost is at this point for the lower end models is around 10 to 20 cents per minute. So you can imagine and by the way that has been improving steadily. So when you sort of back then I drew a chart of cost per minute of video and when you see 2020ish when there was no good AI option when you had to do it manual or pay a freelancer it was around $10 per minute or higher. Now the industry is essentially around $1 per minute and the curve is just going down and down and down. And Moore's law obviously is helping with this. So I think when you just look five 10 years ahead I think the cost of AI video will be even lower than today.

So we're doing, so essentially for our videos we're using... to do good videos for our users we had to integrate a lot of capabilities. So we have AI voices, some of them are third party voices. Some of them are our own hosted open source models. We have obviously these video models. We have...

So for example previously we used and I think still we use on the cheaper end we use LTX Video also. But what we did is it's an open source model. What we did is also further improve it, right? And that's what makes it better fitting to the use cases that we have and for our users. And then we have music, we have subtitles, but I think the core what we're doing is the workflow like how to interact with the user and how to make it really simple and then sort of an intelligent engine which we call this video director which is essentially an AI engine that is attempting to do what traditional movie director would do essentially coming up with the strategy describing the individual scenes, defining the dialogues and who does what and that's what we're building that's what we're focusing on.

So I think for your sort of use case where you're recording a live session, obviously one use case that comes to mind is adding special effects like video B-roll for example or voiceovers. If you for example wanted to make this video accessible to Spanish speakers or French speakers or general...

We support I think three dozens of languages now. We have like 170 voices and three dozen languages.

We do have Chinese, Japanese. We don't have yet... we have Hindi. But yeah so that's essentially we can translate our Neo Show to 30 languages now.

Yeah. I mean, it's all about convenience and accessibility and being able to...

And AI can help with a lot of these things.

So it really depends kind of where on the market map you look for different players. For us where we're spending a lot of our time are essentially three things. One thing is again I don't want to over repeat but the workflow, right? How do we really remove the complexity of the process of taking you from an idea to a video that is ready to be published. So that's one part of it and that comes a lot of thinking, a lot of simplicity and a lot of automation into it. Second thing is the I would call it the consistent character and consistency of time and scene. So if you're let's say you're doing just a four five second short video clip, pretty much doesn't matter and there's a person in the video. You just need to make sure it's consistent within the 5 seconds. But let's say you're doing a 10-second educational video or a 10-second story about two friends going somewhere and discovering, I don't know, Paris. You need the characters be consistent across the videos have some intelligence about the characters to adjust to each scene and to each time period but also have logic flowing through the video so that it looks like a story that has a consistent beginning and end. So that's the consistency and the long form piece. And the third piece where we spend a lot of calories is on being able to support the infrastructure and the deployment. A lot of these models are computationally very extensive. How do you support, how do you scale, how do you deploy to your GPU cloud in a way that is cost effective and that can scale when there's more users at 5:00 pm and there's few users at midnight.

So it comes to the core problem that we're solving which is creating a platform for people to be able to share their expertise or their insights or their unique stories. And then there's just a limited space that you can do it in like five 10 seconds. I mean you can do videos right but if you're like a bunch of our users or most of our users who try to do like CS content then you need a version of long form.

I would say the way we talk about long form is anything above one minute but most of our users are in the five to 15 minutes range. And that's the hardest thing that is out there other than the foundational models for now. So that's what we're solving for our users because we see that as our unique thing that we're bringing to the ecosystem.