The image generation API was just released by OpenAI. We're about to see a gold rush. Like we're about to see a lot of people make a lot of money very quickly. Ads are the perfect output for image generation because they make people money. These marketers, they will pay you a lot of money. It went from unusable but cool to usable really quick. Those things are going to do so many millions of dollars in like the first couple days. It's crazy. What are the biggest opportunities in AI image generation today for someone that has nothing but internet connection and a laptop? There's like a million apps that can be built like that. One of those is going to do incredibly well. All right, I have a super interesting guest today. His name is Jacob Posel and he flew from Australia to San Francisco to build a company inside a company. So, he met a guy on Twitter that owns a software business in San Francisco and they want to build AI tools. So they hire Jacob to build tools to fix the problems that they had within their companies, but also wanting to eventually scale and sell that tool to other companies as well. So this guy is an AI hacker at heart. He's especially good at Chad GPT's new image generation tool. And as luck would have it, the image generation API just became available today. So you will be hearing business ideas for the first time with technology that was never available before today. I got so fired up, I literally had to buy a domain name as he was talking to me. And I'm going to build at least two different software tools solely based off this conversation. This new image generation API is a defining moment in ChatGpt's history. It will enable so much entrepreneurship. So, if you want to hear from someone on the cutting edge of this, we talk about how to build an agency, how to serve ad agencies, how to build chat GPT rappers with image generation at its core, you're going to love it. Please share with a friend. Let's get right to it. What are the biggest opportunities in AI image generation today in your opinion for someone that has nothing but an internet connection and a laptop and a lot of time? There's two ways to take this that can create a lot of money and like okay so today is the 23rd of April and the image generation API was just released by open AI and so we're about to see a gold rush like we're about to see a lot of people make a lot of money very quickly and so I would release this quickly so that these people can take advantage of that. The API was released today and there's lots of different ways you can take advantage. So a lot of people who are not technical and who are not nerdy and who are not on Twitter don't know what chat chat GPT is capable of. I was at an event with some friends recently and we like took a group photo and I was like, "Oh, I'm gonna like giblify this image," which is like a super common thing. And they were like, "I have no idea what that means." And I showed it to them. And you're in San Francisco talking to people your age, right? Yep. Exactly. They're in the finance world, not in the tech world, and they just they just didn't know. And there's like a million apps that can be built like that that take whatever whatever you know everyone is loving that Chad GPT does and just turn that into a super focused app like Giblly Me or whatever you want to call it or like anime me. Mhm. Like one of those is going to do incredibly well. And so you can charge for it based on like you can you can figure out the the math to make it make sense where you sell like a subscription and then maybe credits on top to generate more images. Hope for some virality using a tool like Tik Tok or something like that. And like those things are going to do so many millions of dollars in like the first couple days. It's crazy. Like it's that's coming. And there's a lot of other really cool use cases. Like one which gets people really excited is you can combine two people and be like what would my kid look like? M you know like there's so many Yeah. It's like I did that with me and my fiance and I took a photo. I was like what would our kid look like? And we loved it. It was like we were like I joked that this is going to solve the fertility problem. You know that people are going to want to have kids once they see these. We're all narcissists, right? So if we want to solve the fertility problem and chat GPT has to make us look better than we actually look. And then we're like, "Oh, that's a beautiful baby. Let's have some kids." and the kids do look super cute. Like it doesn't matter who it is and stuff like that. Like be creative there. Like see what like just like off the top of my head like what am I going to look like when I'm older? That's like another one that you can do when it comes to like feeding into people's narcissism and arrogance. Like people want to look at themselves like what would I look like with different hairstyles or with a beard? What would I look like if I lost 30 pounds, right? I mean twothirds of Americans are overweight. So there's your market. Amazing. That's an amazing idea. Yeah. There's like so many of these like feed into people's narcissism. People love looking at versions of themselves. People love taking photos of themselves and seeing what you look like in different under like a different context, like a different in a different environment. So, that's like something that is a no-brainer. There's also like think of other industries where imagery is important. So like like for retail for real estate photography, a really cool one is like take an empty room and like fill it up with furniture for me and like help with interior design. Another one is like improving the lighting or the vibe of like an air like an Airbnb. I posted about that one and people hated it, but there's got to be a there's got to be like an ethical way to do that. But yeah, there's like there's so many of these. There's so many ways that we interact with images in our life and any of that is going to be very important to people and we'll use this API. So that's like the consumer virality side of it and I think we're going to see a lot of really creative people come up with a lot of re really creative things. Jacob, I have to apologize because I told you before I hit record that I can't not launch businesses. So when I said that, you know what what would I look like if I lost 30 pounds and I got your validation that you liked it. I literally just clicked purchase on rightsizeme.com. That's a good domain, isn't it? Amazing. Amazing. Rightizeme.com. How much was it? Done. 10 bucks. It was open. Amazing. I own it. But like that's totally something that people would would pay for and then they would share and then like that literally wasn't available yesterday. Like it it would have been very very hard to build that yesterday, but today the API came out and I think it's hilarious because six days ago I posted a tweet showing like pictures of homes in Texas of like what would this home look like if it were painted white? What would this home look like if a if a large tree fell on it? And I owned a tree trimming business and I wanted to send personalized images of people's homes with a tree on it to incentivize them to pay for my tree trimming services. And half of the negative comments were like API's not even available. And I'm like, "Give it a week." That's what I said. I'm like, "Give it a week." And here here it is 5 days later and it's available. Like people look for so many excuses to not launch. They like, "Oh, there's an excuse. Yeah, I feel better about not launching now." Yeah. Yep. It's so true. People look for a million reasons not to do something, but if you had been preparing for it, you you could have created like a really really great outcome for yourself starting today because it was launched today, which is which is pretty amazing. Have you seen any image generation rappers or tools? I know it just came out today, but have you seen anything that just started printing money right off the bat regarding image generation? No. I mean, unless you count JPT, but like it literally was launched, I would say two hours ago. Like literally two hours ago. So, it's like too early to have seen anything that could have launched. What about like with Replet or the other a image generation APIs? the OpenAI image generation API was like a chat like the GPT4 moment like if people are familiar with that when it comes to image generation it went from unusable but cool to usable really quick and so none of the other APIs were were viable to do something like this and these ones are viable to actually to get past that like churn point you know like a lot of AI apps have really high churn if it just can't get past a certain point and we're there now and So, no, I haven't seen any, but like I guarantee we will see a lot in the next coming days. Tell me about what you were playing with a couple weeks ago with these ads like with the image generation tools and now that the API is available, what businesses do you think people could start with what you were playing around with on Twitter? Yeah. So, this is a big one, which is ads are the perfect output for image generation because they make people money. Mhm. And so I had been playing around with this idea of AI ads for a long time before the image generation API before the image generation chat GPT was even a thing this like new update. And then when it was like long story short it was like the first thing I tried because it's something I I had obsessed over and failed at for a long time before and I had this like wow moment like it's there, it's here, I can finally do it. Um, and I started posting about it and that kind of got a lot of traction, a lot of people impressed. And so I think we are going to see a big influx of AI generated ads. And there is a very high demand in performance marketers to figure out how to use these ads, how to use AI to generate more ads for them because to like simpl oversimplify it, the ad algorithms are starting to reward volume and choice more than high focus. And so a lot of these marketers are trying to figure out systems to create a lot of high creative output at a low cost. And so the market is moving towards that right now. There's a I can say that with absolute certainty. And if someone could figure out how to make it work for these marketers, they will pay you a lot of money because a single ad unit can make them if a single ad cost them a h 100red bucks, they could make like $20,000 from that. You know, like it's the easiest return on investment for them ever. So that said, it's not going to be as easy as people think it's going to be. There's a lot of best practices to be figured out. There's like there's this whole concept of prompt engineering when it comes to regular AI applications, right? Like you can't just go to an AI to an LLM and say like do this task for me. That's not the way you do it. You know, there's a specific way you structure it. You give it a role. You give it an output format. You have to give it the right context. So that similar concept is going to be applied to image generation. How do you optimize the output of images? And how do you create a system and a process to create the highest proportion of viable ads and as far as I know that is currently unsolved. And so if someone wants to create a lot of value here for themselves and capture that demand, I would be focusing all my time on figuring that out. Otherwise, you're going to have a massive spike and a massive churn. And maybe you'll like do a little bit well in the beginning, but you're not going to create valuable long long-term business. Yeah. What are some best practices for prompt engineering images that you found? Do you use reference images? What like do you ask it to do it in a certain style? Like what are your go-tos? Yeah. So, an image is worth a thousand words. Same thing applies here. If you can provide a single reference image, that gives it the AI so much more context into what you want. Explain to everyone what a reference image is or an example of one. First of all, you can you can give it as many images as you want. That's something to that's something to take into account. Specifically, when it comes to ads, a reference image might be an ad that you want to reproduce. So, you have a company, a competitor, and you really like one of their ads, but you want this applied to your brand. So the reference image can be one the reference ad which is just a screenshot of the ad that you like and two can be your product and then you can get the AI to combine those together and turn it into an ad for your product. So just clearly tell it that the ad is a reference ad to reproduce and the image is your product image that you want within that reference ad. Just be clear about that. that will improve the quality of the output like orders of magnitude more than if you were to try and describe that image to the AI. Gotcha. Okay. What's a like an example of doing that? A great example would be you have a lot of people try and a lot of people when they try and make ads they take a winner and they want to create a new variant of that ad a diverse variant of that ad because once an ad has been inside of a inside of an ad network for long enough it becomes fatigued. So you take a a winning ad and you try and create a variant of that. So, a really easy example is take that winning ad, give it to give it to the LLM and ask it to create a new variant, a diverse variant of that ad. Even better, give it to Sora and get it to create four images, four variants. I guess I'm kind of getting ahead of myself here, but like the API is going to let you create like 50 variants of that because now the API is available. But yeah, so you're giving it a reference, you're getting a new variant of it, and you can pick the best one, and you can use that to refresh your account. What is the benefit of using Sora for image generation versus just within chat GPT? So this take might be outdated with the API coming out, which is something to consider, but a couple I was going to say like most people aren't going to use the API that are listening to this in my opinion. So maybe not. An app will come out I think that does things better than Sora is the only reason why I say that. So like Sora is limited to creating four images at once. With the API you can create more than that. But with chat GPT you can only create one image. So Sora is immediately better because it creates four images in parallel which is which means you have more choice when selecting which one you want to use. So store is great because of that functionality because the images it's unlikely you're always going to get the output that you want first try. it's going to take a couple of attempts to do it. Sora is also great because it has a a library that you can reference of like top ranked images and you can click on it and you can see the prompt and that's really amazing for for like learning and for seeing what's working out there and for getting ideas and this is going to be this is unconfirmed. This is just a little personal intuition. And I could be completely wrong, but I think what happened was OpenAI had two versions of their image generation model. They had like the main one which was initially released and then they had a smaller one. It might have been like an earlier checkpoint or like there's a lot of different ways that these engineers can decrease the size of their models, but that reduces quality. It it it reduces compute demand and it lets them it lets them output the images at a cheaper cost, but it reduces quality. And I think that they switched their chart GPT model to a smaller one, the one that was used in CH GPT. And a lot of people noticed that the output quality had decreased, but I think they kept it in I think they kept the original one in Sora. When I was like testing GBT versus Sora with the same prompt, the Sora outputs were a lot better. And so I think Sora does create high quality images which may be just like may be wrong but it's just a little bit of intuition that I have about it. There's just I mean there's just no point to use chat GPT for image generation. You you can move at least four times faster. It likely maybe not definitely but likely has a lower image quality output. So you might as well use Sora. Like if you had to choose gun to your head, you can only generate images on Sora or you can only generate videos on Sora. What do you think is the bigger opportunity? Which would you choose? Images for sure. Videos. Videos still haven't had that moment we spoke about before. They're still unusable. They're still cool, but unusable. It's a gimmick. So, yeah, exactly. It's fully a gimmick. So, video is cool, but you know what's better? long form audio via podcast and my newsletter, tkopod.com. Go there to subscribe for free to my newsletter. It's one email a week, very tactical. And then go to my audio podcast. Three episodes a week, stuff like this. You're going to love it. All free. No sleazy sales tkopod.com. You could think of it like this. If you use if you use text, like I would say 90% of outputs are going to be good. If you use images, 60% of outputs are going to be good. If you use video, 5% of apples are going to be good. So that's until video can get to the that certain threshold where it's useful. I wouldn't really invest a lot of time there. Let the let the guys with like a ton of money training these models invest their time there and when the opportunity comes, take advantage of it. Do you think there's an opportunity for people to kind of surf the the Sora library, sort by most liked, most viewed, and think of like rapper ideas or business ideas based on what those prompts are? Because for those that haven't played around with Sora, like it will show you what the prompt was exactly to generate that image. Do you think there's anything there? Yeah, there's so many ways people get can get creative here. Um, Sora is a great place to go to just like get the get the brain juices flowing and just thinking about what's possible because people are so creative. They're coming up with the most creative things. I highly recommend anyone listening to this just scroll through Sora and just take a look at their explore page. Anything that was like a pre-existing filter that people used, I would take a look at. There's so many ways to get creative here. It's just unbelievable. I'm really excited to see what people come up with. Yeah. Yeah, like I was looking at the the explore page and there was a picture of like Dumbledore, Hermione, and Harry Potter in the gym. They were all like jacked and it was this black and white and it had the most likes and I'm like, you could build a rapper out of that out of that one prompt, right? Because there's a billion people that love Harry Potter. There's just so many ideas. Is the prompt the most important part about all this or is there something else that that people are missing? So, it's task dependent. If it's creating ads, it's the reference image for sure. But okay, think think of the reference image as a part of the proc. It's all the it's not actually truly separate in the way that the AI ingests it. It can take two different modalities and use them both to create an output. And so if you are trying to reproduce something, then the image is the most important thing. If you are trying to create something completely new and creative that doesn't require rep like reproducing something real, then the text, the prompt is the most important thing. Okay. What's an example of when the prompt would be the most important thing? No reference image necessary or as relevant. So like for example, if you're trying to create if you're trying to create that like Harry Potter scene, right? If you're trying to create something truly creative and truly novel or like something like a fantasy using characters or themes that are very very wellnown on the internet. So like for example Harry Potter is very well known on the internet and so the AI already has a very good understanding of what that is then no reference image is necessary like by providing a picture of Harry Potter you're not really moving the needle because you're not going to move the needle. Yeah. Yeah. Exactly. And so in that case it's the most important like for example with creating like a theme like using a theme like the Giblly theme like the prompt there is the most well actually no because you're providing a reference of yourself but yeah it's it's difficult for me to think of a use case where you're not providing a reference image because everything I've done includes reference images because it's all some kind of edit of an existing image that exists. I think there's going to be a lot of winning in like stock imagery here as well. Like you can create any stock image that you could possibly think of and so that wouldn't require a reference image. That will just be a font. Yeah, it's it's super case dependent. Do you think there are opportunities for people that literally their only skill is just knowing their way around Sora and using reference images and using good prompting? Let's say that's all they know. What would be your advice for someone like that to go start an ad agency or to go do like a permissionless marketing strategy and go find like a like a Casper or a DTOC like a Ridge wallet brand out there take their ads and those might be bad examples because they're probably on the cutting edge of this but you know what I mean any company and like without permission make gorgeous ads for them send them to them in a cold email and like start charging like how would you suggest someone approach that? What would they charge? like how could they reach out? What what are your thoughts on that? If you had to start that tomorrow to pay the bills, what would you do? I would probably partner with existing ad agencies and I would say I'm going to help you to offer a service to your customers. You already have customers. They want this. Let me just just whitelist my service and I will I have the skill set to create high-quality images and I have the system to create high quality images and I'll do it for you. And I think if you DM'd any of these agencies on Twitter, they would probably give you a shot because the way that their business is run is it's all it's all like arbitrage on another service. Like it's either labor arbitrage or it's like software arbitrage or it's like another service arbitrage. And so if you can deliver the service to them and they can sell it for more, then they're going to say yes. And so I would just find as many as you can. I would offer this and I would figure I would dedicate my time to uncovering those best practices and understanding what works and what doesn't. And so you'd basically I like to call it piggybacking. You'd piggyback off someone that already has a whole portfolio of not just customers but trusted customers. They already have their foot in the door. They're already paying them. May even if they're dormant customers, they could reactivate them with this. How do you think you would charge? Like would you just say I'm going to take a third of whatever you charge? like you white label this to me like here's how you would sell it here's how you'd promote it here's some examples outsource it to me and we'll split the revenue 50/50 like what do you think that looks like difficult to say the way that ads are usually charged in marketing agencies is like per ad like there's a certain cost per ad value per ad and so I would determine some kind of value per ad which is reasonable and then yeah like I would split the profits probably or I would deliver it to them like I would understand what they would be able to sell it at to their customers and then provide a reasonable markup with them and then it becomes a really nice volume game right like you just everyone's incentivized to just create more ads and there's almost no cap that the brand will buy from the agency if the ads are ready to go into the account because it's very unlikely that they're going to lose money on that in the long term and so I'd probably do it by ad or you could figure out like a cost that makes sense as like a base plan, right? And like they just charge you that as like a retainer if they're willing to do that like regardless of how many ads are accepted or not. I think that would really work as well that like is a little bit that reduces your upside but decreases your downside or or do some combination where it's like a base retainer plus number of ads accepted by customers and launch into the account. Something like that. Dude, I like that. Here's another idea that just came to me. So, a lot of these ad agencies list out their customers, right? They just make it too easy for growth hackers like us. You go scrape a list of all these ad agencies, preferably ones that aren't on the cutting edge, aren't already doing this. You build a rapper with Replet that just can basically do what Sora does, right? But you brand it towards ad agencies only. You're not going towards, you're not trying to get brands or companies. You want to go through the ad agencies and then you go to their portfolio page and you see, all right, John ad agency has Athletic Greens as a client. Okay, I'm going to go remake six Athletic Greens ads and I'm going to send them to them proactively with a link to my rapper that I built in Replet that allows them to do this themselves. You get them using it. You get them on the interface loyalty. It's in their workflow. They're sharing it with their account managers, etc. And then you charge 200 bucks a month for it. And it's just a tool. It's just a tool that they use for unlimited image generation. Now, you'd have to watch the credits, right? Depending on how much they use it. But I like that model over like charging per ad because usually brands want to know like what is this going to cost me per month? Like just I just want to know. I don't want to be surprised with a big bill. What's it going to cost me? So really we're just talking about like a rapper for ad agencies. What do you think about that? I think in the beginning you need to do it for the agencies. I think like remove the friction. Yeah, I think tools are good, but it's really difficult to get people to use tools, like to actually use it themselves, even if there's really obvious value there. And like the dream scenario for an agency is they don't need to think about it and they make money. And so, while you can, I would deliver the service for them. Like, I would literally send them Google Drive folders of just completed ads. I would ask them what format their client expects ads in. And I wouldn't overthink it. I wouldn't overthink the software. It would just be like, how do you deliver ads to your client? Okay, I'm going to build you ads and I'm going to put it in that format for you and I'll send it to you and all you need to do is pass it off to them. And I think that behavior of an ad of a brand approving ads from an agency and the agency and then charging based on the number of ads they approve, I think that's normal. And they're not going to be hit with a massive bill, you know, like they could be provided ads and they don't need to accept all of them. Like they're very famili they're aware of how much they're going to be paying for these, you know? Oh man, my mind is like buzzing right now. We I saw you either on Twitter or on another YouTube video talking about micropersonalization with these image generation tools. What What are your thoughts on that? I think it's kind of amazing how marketing can change now where the bottleneck in how you managed almost everything was the creative and like the dream of marketing are these hyperpersonalized hyperargeted onetoone interactions with your customer and so like that's obviously impossible. So C companies think of like different personas and they try and build towards those personas and let's just say based on the bottl based on your creative bottleneck you could only target four personas a month. Right now if you can create so much more creative then the floodgates open on the ability that you can market yourself. So rather than trying to market middle-aged women in America, now you can target upper middle class women who grew up in a certain postcode in southern Florida. You know, like there's like ways that you can really nail down your personas because now you're able to actually create creative for them that targets them at a scale which and at a cost which makes sense for you. I think that like that's going to really change marketing. I'm really excited about that too. You know, like I spoke to an events company that runs events and it was really difficult for them to market every single event because how are they going to create an ad for every event and how they going to create an ad for every person who might go to an event, you know, and that just wasn't possible, but now it is. And so there's a lot of different ways that marketing is going to change. And I'm hoping it results in higher quality targeted ads so you have less just like slop on your feed. But it might start the other way first. But it's exciting. It's exciting to where it could go. Okay. So, you had mentioned like a rapper that takes two people that makes them shows them what their baby would look like, right? Yeah. I'm going to share my screen. I just did that. Are you ready? Please. So, I went, you don't know what my wife looks like or my kid looks like, but this is my prompt. Okay. So, I took a blurry image. Like, this was like it was very blurry. I just screenshotted it. That's me. Yeah. And that's my wife. And I said, "What? Such a cute kid. What would this couple's baby look like?" Okay. And you see the baby, it's a cute baby. Yeah. And then I went to my Google Photos and it looks like about like a one-year-old and it reminds me of my youngest son and I got this right there. You see it? Yeah, I see that for sure. It's weird. That's crazy. I'm going to build a rapper that does this like by the end of the week. Please, please do it. Please. It's so It's so good. I love that. How do you stay focused on this one project you're working on with all this crazy stuff going on? It's tough. It is really tough. But I just know where my strengths and weaknesses are. And when I get to spread thin, then I kind of don't really win at anything. Yeah. And so I've made the very conscious effort to drop everything else and only focus on one thing for better or worse. You know that I may end up regretting that, but I'm going to give myself that opportunity for now and see where that takes me. I love it. Well, Jacob, this was amazing. Thank you for your time. Where can everyone find you if they want to reach out? Follow me on Twitter, Jacob_posal. I also post on LinkedIn sometimes if you want to follow me there. Uh yeah, if anyone builds anything cool, please share it with me. I would love to see it. Love to take a look. I really enjoy talking to people who are creative and who are building and who enjoy building cuz that's my obsession. So if that's you, please reach out and just we can we can talk through these things. Okay. Thanks, man. Appreciate it. Okay. Please share with a friend. Leave me a fivestar review. I would greatly appreciate it. and we'll see you next time on the Coloner