Meet the AI Meme Millionaire

0:00This is not your standard SIP episode. I brought on this guy who went from zero to 250 million mobile app installs using some open source AI models. He tells the whole story, how he did it. You know, it definitely fired me up around this whole new wave of creating, you know, multi-million dollar mobile apps in under seven months.

0:33Ben, you've created these AI apps. You've got hundreds of millions of downloads. You're backed by NVIDIA. I want you to teach us. Well, actually, what do you want to teach us before I want you to teach us? Yeah, I thought I'd just come on here and talk about a strategy that's worked well for me in many different domains in life. And it's sort of a basic fundamental of mimesis, which is if something is working then probably you can copy it and put your own spin on it and it'll continue to work okay and by the end of this episode like what are the concrete things that people are going to learn i'm going to show you how i've applied that strategy to a few different uh successful things in my life how i used it to come up with a concept for an app my first mobile app ever that got over 100 million downloads how we use that concept to consistently go viral on instagram and have reels that get millions of views and that are on one hand a remix of something else that already exists but on the other hand something that we put our own unique flavor on okay i mean how to create apps that get millions of downloads how to get how to create content that gets millions of views, how to raise millions of dollars from people like NVIDIA, you have my attention.

1:56Okay, let's do it. It's actually pretty fucking easy if you don't overthink it too much. So, okay. First, very brief anecdote. I think I picked up this principle from video games. I was a jungle main in League of Legends. I was sort of like a platinum gold level player, which is like 60th, 70th percentile. I wanted to get better. So I started watching streams of players that were better than me, that were playing the same role, and I would just copy them and then put my own spin. So that's where this sort of entered my head for the first time. In the summer of 2020, I applied this to conceptualize and eventually launch one of the most viral apps ever, Wombo.

2:37So there were a few signals that I picked up on. first was Reface. Reface in the summer of 2020 was the number one app on the app store and Dreesen invested and led a five and a half million dollar round. And what got my attention about this was it was an app that I fully understood. So what did Reface do? Reface did face swaps. So you could put in like a Cardi B music video, take a selfie, it would put your selfie into the Cardi B music video. Hundreds of millions of downloads, memes that my friends were sending each other. I knew which model they were using and how they were accomplishing this.

3:16And the design of the app was also very simple and intuitive and kind of made sense to me. So that was the first seed. The second seed was this PewDiePie video where he covered some very strange memes. And we can see the date here, four years ago, August 2020. It's beautiful. It's beautiful. And of course, this got 7 million views because it's PewDiePie. And this got my attention as well because this was another open source AI model that I was familiar with called the First Order Motion model. And it was starting to go viral in sort of the AI circles and also just like the hardcore like Discord artist circles, like people on their computers for like 14 hours a day.

4:04People like us. Yeah, people like us. and they were using a Google collab notebook to basically stitch this output together. You gave it a selfie. You gave it a driving video, which is what kind of creates this animated choreography. It's actually someone else moving their head around. And so this one driving video of a guy singing Baka Mitai went viral over and over and over again. And so one night I'm smoking a joint with my roommate on the roof. We're looking at these memes. Someone had made a Telegram bot where you could do something similar. we're like man like why doesn't someone just make an app like where anyone can do this where it's simple uh and so immediately like it kind of starts off as a joke but the circuits start firing like oh wait like reface just did the same thing they have hundreds of millions of downloads here's a new open source say i model that like no one's done this for yet like easy like we're going to be billionaires we'll get this done in a week you know um did not take a week took us seven months but in march of 2021 we launched wombo and maybe i can find a quick uh walkthrough on youtube when when you had this idea and you saw that a16z reface company like a lot of people would look at that and be like okay they're gonna win this space like you know it's already gonna to get competitive but you kind of said no no i'm gonna create an i'm gonna compete with them yeah i wasn't thinking that i'm gonna compete with them i was just thinking like it's early like there's lots of space to play like i don't know like there's where my mind goes to is like like like movie directors or or musical artists like we're never gonna run out of like new movie directors or artists like we're going to continuously want new interesting shit so i don't think anything is like too crowded yeah abundance mindset yeah sam altman the co-founder of open ai just said that it is the era of the idea guy and he is not wrong i think that right now is an incredible time to be building a startup and if you listen to this podcast chances are you think so too.

6:21Now, I think that you can look at trends to basically figure out what are the startup ideas you should be building. So that's exactly why I built idea browser.com. Every single day, you're going to get a free startup idea in your inbox. And it's all backed by high quality data trends, how we do it, people always ask, we use AI agents to go and search, what are people looking for? and what are they screaming for in terms of products that you should be building. And then we hand it on a silver platter for you to go check out. We do have a few paid plans that take it to the next level, give you more ideas, give you more AI agents and more, almost like a chat GBT for ideas with it.

7:06But you can start for free, ideabrowser.com. And if you're listening to this, I highly recommend it. So I'll just kind of briefly walk through this. you can actually see that in this video this lady shows this old google collab notebook for the first order motion model it's like my grandma could never use this shit like i mean she could if i like taught her python yeah but like she could never use that um so here we go uh you would take a selfie that's the first thing you do then you pick a song and then you sit through a little loading screen like this. And then finally you get your output and we'll just see that at the end.

7:50Obviously it looks super basic by 2025 standards, but literally those three screens of input, song selection, output, or sorry, loading screen and then output, four screens, that's all we needed to get 100 million downloads. And we intentionally sort of designed something that would be extremely simple to use. Sort of the whole thing is engineered for virality in a way, because the whole expectation is you as a user come in, you make a piece of content, you make that piece of content extremely easily and extremely quickly. And then your natural inclination is to share or to make another one where like your dad is the one singing or the Pope or a politician or whatever.

8:39And so we spent no money on marketing and spent all our money on inference. And that's how the thing went viral. Maybe the only other thing I'll say here, and I'll kind of kiss my own ass a little bit here, is the song selection. We selected 15 songs that we launched with. And these are some of the most recognizable and iconic songs of all time. or there were songs, there were kind of meme songs. And so like, you know, one example of something that would make a song, a meme song is you don't need to know English to vibe with it. For example, uh, boom, boom, boom, boom by the Venga boys. You don't need to know English.

9:24No. Sounds great. Yeah, absolutely. Wherever you are. Yeah. And so we, we really deliberately picked those 15 songs and then we really deliberately made the 15 driving videos for those songs. and uh that was i think key and and you made it like stupid simple stupid stupid like this is you can give this to a five-year-old and they'll know what to do exactly a buddy of mine has a an app that's going viral right now maybe you've seen it mixy mixy yeah have you seen that i think so yeah so it's like a mashup um like go to just just google mixi mixy app i think it's with an i there it is oh yeah i've seen this yeah so it's like two guys um you know you put two beats together it mashes it up like simple idea right but the ui is really simple um and they're getting like i mean it's one of the top music apps um i don't know the monetization of this but i think there's this whole next there's this whole new generation of apps that are just like really stupid simple um that uh i mean it's a similar concept right like i'm sure his songs are like he's doing a queue of songs that you know people actually want to choose right um it's simple ui um and it's viral because you want to share that on ig etc yeah i i personally like starting with Stupid Simple.

10:59And maybe from Stupid Simple, you can do something more advanced and interesting. With respect to mobile apps, right? With respect to mobile apps. And I mean, it all kind of boils down to attention at the end of the day. Like if you're trying to make a product, it's not just someone's money that you're after. Like even before that, it's their attention. And there's so many fucking distractions. There's so many different things that are pulling you in a million different directions at all times that I find the best way to sort of introduce myself and get someone's attention for the first time is to do something stupid and do something funny, do something simple, not try to fucking explain Socrates to them or something.

11:40Um, but just basically tell a small joke or another analogy could be like, give them a small piece of candy as opposed to like, you know, here's a fucking steak. Right. Yeah. Cause what I'm trying to figure out, what I'm trying to reverse engineer is how do you create a mobile app that gets millions of downloads so stupid simple one what else uh so uh in our case there's always been some kind of viral content creation loop meaning someone comes in makes a piece of content shares that content on other platforms and that's what drives your user acquisition um and so that whole uh that whole component has a lot of extremely important details.

12:24So first is the content creation. How are you going to help someone make a good piece of content? Also, the game has changed since five years ago with how far generative AI has come and the kind of content that people are interested in making. But still, I think it's applicable today. So here's how I would boil it down. How do you help someone make a piece of content that either they're going to post on Instagram and post on TikTok for the algorithm and the algorithm is like that's a great piece of content let's give it millions of views or more for their own like personal networks and the people in the personal network are going to be like wow that's really awesome i want to make that thing too so you're basically saying you start with the content like the end content yes which is kind of what my friend joe is doing with mixy right because like he knows that people want to share these mixes absolutely and so like Like, you know, one thing maybe we could touch on briefly is just the algorithmic feeds of today.

13:22Like the amount of human attention that is just being sucked into the full screen, infinite scroll format, whether it's TikTok or Instagram or YouTube shorts or, you know, X has one. Like that is just like the battlefield of attention. And so I think like a great source of product inspiration is to look at that and look at where are people's attention going? What kind of content are people making? can I make it easier for them to make some piece of content that is consistently going viral over and over again? And then put my watermark in the bottom right. Does this only work for B2C fun apps or could this also work for B2B or prosumer type apps?

14:04I think it could work for everything. You probably need to adjust your principles a little bit. and like you know maybe i'm gonna stop paying attention to the tiktok algorithm and start paying attention to like i don't know the conference that all the people in a particular niche are going to but it's the same thing like what do the people you care about pay attention to right and where right yeah and and sort of the key the key point is like look at what other you know Look at the content, look at what's going viral, and then apply that to whatever it is you're doing. Yes. So in the case of Wombo in the summer of 2020, it was, well, Refa has just got a number one app and hundreds of millions of downloads.

14:51And hey, there's this new model now, which is still very hard to use. This is another very good signal, by the way. if something is going viral inside of like some technical circle, but it's still difficult for a layperson to use, then if you can help the layperson get access to that same technology and make it easier for them, that's typically like a really good recipe for virality or a really good sign that something might go viral. So I'll just wrap up the Wombo story. from kind of that moment of inspiration. It took us seven months to actually get the app on the App Store. But we finally launched it in March of 2021.

15:34It went insanely viral, got like 50 million downloads in its first month, in the first three months. And, you know, after sort of some just like stabilizing and taking it all in and raising a little bit of money and building out our team, the obvious like next question was, okay well we just did this with one ai model what other ai models can we do this with and now this is a really obvious question to ask in 2025 in the age of ai um again 2020 was still a little bit different earlier generative ai was not even a buzzword yet and generative ai is already a buzzword that we've moved past now why what's the latest now i feel like now it's just like agi super intelligence like agents you know uh so the next concept was dream which was an artwork generator um for the nerds that predates mid-journey and stable diffusion and dali and was basically the first time that image generation went viral now of course image generation is everywhere um and uh we saw something very similar happening again almost the exact same thing we saw happening with Wombo.

16:45People were playing with a combination of two models, VQGAN. Fuck, VQGAN and Prashan, I need your help. I forgot. What were the two models? Huh? It was VQGAN diffusion? VQGAN clip. That's right. VQGAN clip. So you combine VQGAN with clip. You could generate these crazy dreamlike images. There was like maybe 20 or 30 people on Twitter and really like five at the beginning who were like messing around with this stuff. And those were like the earliest incantations of like this era's neural networks for image synthesis. And we did the same thing. We wrapped that up into a fun and easy to use mobile app.

17:31We made an extremely optimized version of the model, which had fast generation times. And we did some templating to make it easier to create high quality images, basically just adding words to the prompt that the user gave us. And exactly the same thing happened. Another number one app. And many of the lessons that we learned from Wombo about give the user a simple to provide input. In the case of Wombo is a selfie. In the case of Dreamo is a text prompt. Pre-process that user's input with your own input that kind of like structures it and makes it work well. So in the case of Wombo, it was the driving video.

18:18In the case of Dream, it was the kind of stylistic keywords that we were adding to the user's prompt. Make the inference really, really fast and free for the user. Did that in both cases. Put all that together in a fun and easy to use interface and you have something with like all the ingredients for virality. okay and um like okay so it goes number one what next yeah so just in terms of the trajectory of the company listen i had no idea what the fuck i was doing like i'm not like one of these like silicon valley kids who like watched every paul graham video i was just like smoking weed in my bedroom playing league of legends and like really really interested in ai um so So made every mistake we could.

19:10But at that time, now we're sort of at the end of 2021. We had a team of about 20 people all working here in Toronto. Got a few apartments that we were using as offices. In 2022, we're like, let's raise money because that's what you're supposed to do. ended up putting together a series A where the lead investor kept us in due diligence for two months and then pulled out of the deal. And they pulled out of the deal partially because our original Wombo app used music and there were some sort of unknown liabilities and risks around our use of music. I don't know if I should bring it up with these Mixi guys, but fucking the music licensing people are vicious, annoying, motherfuckers but also you know it is what it is um and also partially because ftx crashed and uh all this crazy shit was happening in crypto and there was over yeah there was like a market spook so we went from having fucking number one app thinking we're about to get 15 million dollars to having negative runway and being like what the fuck just happened negative runway meaning like you're out of money yeah like or you're like really out of money like really out of money where and it really happened because that month dream went viral again and we had a million dollar server bill on amazon for that month for all the influence that we were doing and i had like maybe not the perfect philosophy around this which which was like money is fake let's just make an amazing fucking product let's get all the users we'll figure it out later um Um, yeah, didn't work at that point.

20:58So, uh, we went into like a year and a half of cockroach mode, basically, where I wasn't paying any of my bills unless I absolutely had to. We were extremely focused on monetizing our products, which previously we had never given a shit about and basically getting to like profitability. And along the way, we, you know, still raised some money from investors who believed in us. uh so yeah about a year and a half of cockroach mode got to profitability and how did you monetize the apps because i think because i think that's a question that a lot of people have which is like okay great now i've built this mobile app it's starting to get some traction yeah what you know how do you think about monetizing like what what's the sauce around monetizing mobile apps just copy people who are better than you so who's the best who's the best and what are they doing fucking nikita man that guy is the that guy's a great monetizer of apps

22:00shout out nikki to beer yeah go oh i guess you can't do it anymore but go like pay ten thousand dollars to talk to him on intro and i'm sure he'll give you some good tips okay so how did you how did you monetize it uh we introduced a subscription then we introduced ads the subscription did things like well if you want to generate an image you can generate four at a time instead if you're a subscriber or you could have access to some premium styles if you're a subscriber and that took us from zero to I think at our peak Dream was doing over$500 ,000 a month and that got us to the profitability and sort of stability that we needed.

22:41And what are we talking about? Like$5 a month,$5 a week. I know a lot of people are doing weekly subscriptions now. Like what was it? We've done everything from$5 a week to$5 a month. We started really cheap and got more expensive over time. And there's also like all the dark patterns that you could do with like hidden trials and all that kind of stuff. We never went too deep into that. Yeah. Okay. I'm sure you studied Cal.ai and what they do yeah yeah what do you think? it's a new generation of killers man I love to see it it seems like every few months now there's like a new 18 year old kid making 2 million dollars a month on some simple stupid shit and I think it's only going to accelerate yeah and those apps are monetizing via subscriptions right?

23:38weekly? monthly? i'm not sure we'd have to yeah double check i think weekly yeah i think a lot what i that's a trend i'm seeing a lot is like weekly subscriptions anywhere from like three dollars to twenty dollars a week make sense yeah yeah cool okay so going back so you're you start making money basically and then what next yeah so we slowly emerge out of cockroach mode i guess bringing us into 2023 and 2024, the most notable things that happened were, one, we had a relationship with NVIDIA. They approached us very early on because they saw that our apps were going viral and that we were doing this thing that was going to be called generative AI.

24:26And they want to kind of learn more about it and see how they could support it. Of course, it's evolved like far, far beyond that now. And that culminated in us meeting them more and more and them making an investment into Wombo. The second thing that happened was I sort of, I lost some interest in making apps, and I gained more interest in, you know, where all of this was going, where AI was going. Oh man, this is going to be kind of a convoluted point, but it'll make sense at the end. Okay, so one thing that we've always cared about is inference efficiency. How do we make generation times extremely fast?

25:14And how do we spend the least amount of money that we can on powering this inference? So initially, that meant we are going to host our models on GPUs, we're going to sign a three-year contract with Amazon where they give us some better pricing and we're going to pay specialists$3 ,000 a day to make our model produce the same output but in a fraction of the time. And it's just worth it to find a specialist like that. That was sort of the first solution. And then we're going to run all the inference on the cloud. Our users never have to worry about it. And it's consistent. It's high quality. It's reliable.

26:01then there was a motivation to move some of this inference on device, especially as your user base has devices in it that are powerful and perfectly capable of running the model that you care about. And so we started doing a little bit of that. And sort of the natural extension of that is, well, if there's someone in my user base with an iPhone 16 Pro or a computer with a 4090 on it that can run their own workloads, what if they could run workloads for other users as well? And so that idea of sort of peer-to-peer computing and using our users' compute was born out of our continuous push towards inference efficiency.

26:48Hopefully all of that is making sense so far. And so the other thing that happened was at the end of 2023, at the start of 2024, I got really interested in how AI and crypto would intersect. And so I looked at every project in the space, studied them closely, again, coming back to that same principle of copy what other people are doing, or at least understand what other people are doing before you do your own thing. And so we looked at BitTensor and Jensen and Render and Akash and Golem and the internet computer and folding at home and all these other sorts of examples of decentralized AI or decentralized computing happening.

27:34And I just ended up finding that extremely interesting. And to make a long story short, aside from learning about those projects and even building in some of their spaces, the conclusion we came to was to build our own thing called W.ai, which would essentially allow anyone to turn their MacBook, their gaming computer, and in the future their PlayStation or iPhone or whatever into a compute contributor. And so it's almost like Airbnb for that device. When you're not using it, you can make it available to the world. It's going to run some hopefully valuable workload and earn you a reward. And reward being money?

28:15Reward today being points, which will later turn into a token, which will later turn into money. I'm not media trained, so I don't know what I'm allowed to say. You're not media trained? I had no idea. No, I'm not fucking media trained. But the idea is you are contributing to the network. If I let W.ai generate 10 ,000 images for somebody else on my machine, then I should be able to generate at least 10 ,001 images in return. And that seems fair. Yeah, totally. Cool. And that's the journey. That's the journey. A lot of crazy shit has happened in between. Cycled through the team many, many times.

28:56We started with six co-founders. Only one remains. Three of them have started new companies that are very fucking cool and interesting. So check out Omega. Check out Dippy. That's the Wombo Mafia. And now we all live in a big ass house in LA. There's like 12 of us who live and work there. There's a few of us who don't live and work there, but are still core parts of the team. And just to tie a bow on it, I think that there's a huge opportunity to build, and I know you agree too, build mobile apps that have an AI component to it. Like every startup, most fail. If you want to create a mobile app with some AI component to it that gets millions of downloads, the way to do it is step one see how other people see see like understand your landscape what are the what are the other apps go look at cali go look at risa go look at all those apps right study them study like understand that every single screen yes why they're doing it how they're doing it what colors they're using what typography everything step one step two is from a content perspective understand what go is going viral in a bunch of different niches right and and we can we can stay on this point a little bit i have two collections that i think of interest here this group chat on instagram and this collection which i can actually make public and share with you if you'd like this has been me sort of cataloging ai generated content or content that could have been generated by ai which has gone viral on these respective platforms for the last like two years uh so for example let's go there was like a week or two where i was getting a bunch of these baby videos yeah like theo vaughn yeah yeah i i've seen these like the baby this is with hydra right yeah this is with hydra and i think they were using chat gpt to turn the original image into like a babyfied image okay and then hedra to animate it okay hedra.ai for folks who haven't checked it out chat gpt to create the reference image yes feed that to hedra okay yeah so i'll upload like a picture of heisenberg be like turn him into a baby and then take that baby picture bring it to hedra so maybe someone you know for the week or two that this was going viral or even now can make an app that just fucking abstracts that away from people for They don't have to worry about chat GPT, they don't have to worry about head draw.

31:36It's just take a selfie, choose a scenario, done. So it's basically productizing the content format is what you're saying. Yes, exactly. It's also just useful to study how this stuff changes over time because the AI is not just transforming or creating content in one way. It's doing a bunch of different things. so is your advice to people just like scroll and then just save yes if you want to brain rot then you must be brain rotted first right true so lately we've been having a lot of success with some really stupid shit uh and it's churning celebrities extremely recognizable celebrities into pregnant cats okay like the trump let's go open the trump one hey Elon it's time for our litter to be born

32:48so like why does this work it works because you have celebrities who are recognizable works because it's kind of controversial it works because pregnancy is a concept that everybody can evolutionarily like understand and then male pregnancy is just something that shocks your mind a little bit um and it's top of mind right this is right around when elon and trump were fighting yes right now the other thing about that neem i just showed you is we've repeated that exact same scenario, that exact same storyline, like six different times, and it goes viral each time, including this one that has almost 20 million views.

33:32Hey, Messi, it's time for our baby to be born.

33:42Hey, Ronaldo, why does this baby do that? I show speed.

33:49so literally the exact same thing just changing the characters changing the settings a little bit right and as a result like we can pump that out in like an hour and so that that got 18.8 million views on instagram yeah and your goal is to get you know how many downloads from that uh well hopefully like even one or two percent is pretty good yeah um right now it's a weird time for us where we don't care at all about our mobile apps and we care 100 % about W.ai. And this content is not really about driving downloads to W.ai. It's more about keeping us sharp and on top of the algorithm. I think that's really important.

34:33The reason I ask is just for people who are creating content like this, what sort of conversion rate should they expect? Yeah, I think 1 % to 2 % is like a win. And I think also a lot of the downloads are not going to... Like it's possible that your downloads won't come from the video getting lots of views on the algorithm, but from people sharing some piece of personalized content with their friends. So like for example, let's say we turned this thing, like whatever this recipe is, into an app. like if i turn myself into a pregnant cat i might not post that on instagram but i might post that into like a whatsapp group chat and it'll go like that'll lead to some reality someone someone in the comment section i can hear it already now is is gonna say okay i'm gonna go create an app that turns you into a pregnant cat yes i want 10 no no that's not even what they're gonna say they're gonna say and you think that you're gonna get millions of downloads and then you're gonna be able to charge people three dollars a week for the privilege to do that that's crazy you are you guys are out to lunch that's that's what someone in the comment section is saying right now yeah i think uh i think that person is not entirely wrong but they are wrong like if you make an app and it goes viral even if it goes viral once and people never use it again like that's a eight nine potentially nine figure opportunity um think about it like you're making a movie like if you make a movie people go to the movie theater and spend 20 bucks to watch the movie another 30 bucks to get a bunch of fucking popcorn and shit and uh then they leave and they never think about that movie ever again so like you're just giving someone a bite-sized version of that that they paid$3 for instead of$50.

36:32So, yeah, I think, you know. The other thing is with these apps, to get millions of downloads, 98 % of people are going to use them for free, and you're going to make all your money on the 2 % who don't. The 98 % are going to drive your virality and keep making your content and sharing it with their families and friends and getting you to the top of the app store. And then the 2 % of people who don't give a shit and they're just clicking buttons and they want the extra thing that you can't get for free, that's where you're actually going to make all your money. Cool. And maybe let's end it on, for people listening, do you really believe that there's a lot of opportunity to create mobile apps with AI right now?