Ads and AI: Leveraging AI Creative in 2026

There are absolutely advertisers having tremendous success with AI advertising. It's an insane time. Like, I, I'm in this constantly, and I'm constantly overwhelmed as well. What are the risks when you are using a AI-generated image or an AI-generated video of a person effectively representing your product? There are some products I would say that maybe are a little bit more sensitive, mainly more apparel or, like, skincare, like any of these kinds of brands that are a little bit more specific to, like, how an actual person uses it. Do you believe that this is going to become the standard here pretty soon, where most creative is going to be either enhanced by AI or wholesale created by AI? I do. Today I'm very excited to be joined by an AI ad strategist who specializes in AI creative. He's known online as MrPaidSocial. He helps marketers and entrepreneurs improve their return on ad spend with AI ads. His membership is the AI Ad Alchemist. Caleb Kruse, welcome to the show. How you doing today? Doing great, yeah. Thanks for having me. Stoked to be here. Awesome. Well, tell us a little bit about your story. How in the world did you get into, ad creative and ultimately into AI? Yeah. Yeah, awesome. Um, so I've been working in, like, paid social advertising space as a media buyer for the last 12 years. Uh, the last six years, I was working at a company called Fluent, where I was overseeing their whole paid social, like, arm. Had a team of, like, eight media buyers I was managing. We were spending quite a bit of money on, you know, Meta, TikTok, Snapchat, all social media platforms. And so a lot of what my role turned into towards the last couple years was, like, how do I make my team as efficient as possible? Um, so a lot of what I would be doing is building out automations, you know, figuring out AI. At that point, it was like 2022, 2023. There was no generative AI images or videos yet, or if they were, they were pretty terrible. So a lot of what I was focused at the time was m- mainly just leveraging things like ChatGPT for, you know, thinking of, like, new ad copy variations or brainstorming creative ideas or things like that, or, you know, using tools for bulk ad creation or automating certain parts of the media buying cycle. Uh, so yeah, in 2023, I quit my job, for a variety of reasons, but I was just, like, you know, ready for something new. And started making TikToks about everything that I'd learned over the last, at that point, 10 years about automation, figuring out AI along with everybody else. Mm. And then, yeah, I, I think about three months of doing that, kind of cracked the code on TikTok. Started getting a following, and it was clear to me, okay, well, maybe content creation is kind of my little pocket here. I'm just gonna keep going with that. It's clearly providing a business for me. So just leaned into that. And then in 2024, some- somebody posted a video on Twitter of, like, a woman talking about body odor, and it went viral, and everyone was, like, talking about how it looked crazy real, but it was AI. And I saw that, saw that it was kind of going viral over Twitter. I made some videos about it on my TikTok. Those went viral, like 8 million views or something crazy. And then not long after that, the founder of that company who put the video out reached out to me, to see if I wanted to collab on some content. So kind of a fun, like, you know, content creator meets, you know, up-and-coming SAS ad performance tool. Um, so yeah, we, we started collabing on content. Um, and fast-forward to today, that company, they're called Arcads. Um, they are one of my biggest partners, who, you know, we collaborate. They, they compensate me for creating content on my channels. And, I... A lot of what I do for them is when a new model comes out, I get my hands in on the tool, I use it through their platform on Arcads, and I figure out, like, how can advertisers be us- utilizing this technology for creating image ads, video ads, leveling up what they're doing already, but just layering in AI on top of it. And so a lot of my content on my channels is AI generative content related. Love it. Okay, so you, basically worked inside of a company for a while, and you really mastered the, paid acquisition side of things, and then you went- Yeah... and you created content all around, like, your craft and how you did this, and then eventually, as AI came on the scene, you started exploring how you could use AI-created creative, for lack of better words, to enhance, the ads that, that you were creating. And before you know it, boom, all of a sudden you now have this following. You've got your membership called the AI Ad Alchemist, and you're basically teaching people all over the world how to do this stuff. Do I have that right? Yeah. That's a pretty good summary. So, there's a lot of people that are listening right now that are definitely dabbling with ads on Instagram and Facebook and TikTok and all the different

platforms. Um, why does ad creative, that is AI ad creative, matter? Like, talk to people a little bit about that. Yeah. I think, you know, for one, there's a lot of folks out there that are scared of it, and it makes sense. It's kind of like people aren't really sure exactly how they feel about it yet. But the reality is, it's where things are headed. I mean, you look at Meta as an organization, and their whole mission is to completely automate the media buying cycle with AI, or, like, you can just put in your product URL-Um, describe what your product is, give it a budget, and then it, it'll run, and it will eventually create all these ad creatives without you even lifting a finger. They've moved already very close to that with their acquisition of Manus recently, which you have-- if you haven't checked out Manus, it's a fully, like, agentic, tool that, like, will do everything from, like, analyze your media buying to, generate creatives for you and all this stuff. So they're, they're-- like, Meta as a platform is already moving significantly closer to integrating AI. Um, so these third-party tools that have, like, have popped up that are kind of, like, ahead of that curve, you know, there's already a lot D2C advertisers, a lot of, just media buyers that are leveraging this technology today, and it's kind of, it's something that you're gonna have to learn, like, regardless of how you feel about it. And, like, if you wait, then you're gonna have to play catch-up with the people that have already figured it out. Does it actually help with conversions, I guess, is one of the big questions. Like, what is the upside? Um, what are some of the benefits that you're seeing and/or your students are seeing, or that your clients are seeing as a result of leveraging, AI creative? Yeah, I mean, I think bro- like, s-speaking broadly to, like, people in my school community, you know, people that I go on consulting calls with, you know, to looking at just, like, the user base of Arc Ads as, like, a, a reference point as well, like, there are absolutely, advertisers having tremendous success with AI advertising. Um, like, great, great return on ad spend. Um, and I-- but, like, I would say the larger impact there is actually from, one, time saved, so you don't have to have the traditional back and forth with a real UGC content creator that can take weeks or months, depending on, you know, how responsive that person is. You can get, you know, a hundred videos done in the time it might take you to get one final video done from a real UGC creator. The cost per video, like, per finished product, is usually, much less than if you're going and trying to hire, like, a content creator like me or, you know, just an average person to demonstrate your product. And, yeah, it, it overall is just like, it's saving people time. It's letting them expand on how much creative they can put together. And then thirdly, when you think through where the platforms are headed as well with the Andromeda update and, like, creative diversity and feeding more and more creative into these platforms, AI allows you to do that at scale much more efficiently, where you can have, like, maybe the same script that's generated by, like, eight different personas in different settings that hit your target demographic in different pockets that then the system can then kind of, like, find those users based off of your creative signals. Love it. Okay. Um, we're gonna get into how to do this in a little while, but first, what do we need to be thinking

about? Um, what are some of the either mindset shifts or groundwork that we need to process before we actually begin using AI, to create, ads? Yeah. I mean, I touched on it earlier, but, you know, there is kind of the, a brand safety tolerance that you kinda have to figure out for your own, your own business. Like, how comfortable are you having an AI-generated person, in your ads or, like, holding your products or, like, you know, having an AI-generated image? Um, because there are some brand, you know, risks associated with that. I think there's-- that's probably, like, the biggest barrier for mass adoption at this point is people that don't want to look, deceptive to their audience. So next- Well, let's look at that for a second. Like- Yeah... what are the risks, or at least what are you hearing from other people that they believe are the risks when you are using a AI-generated, sh- image or an AI-generated video of a person effectively representing your product? Like, let's talk about that a little bit. Yeah. I think the risks are that, and it varies very much depending on product, de vending-- depending on industry. But, like, there are some products, I would say, that maybe are a little bit more sensitive to this type of thing than others. Like, if you're doing maybe lead gen, and it's, like, you know, something like, solar installs or something that's just, like, there's not really a lot of brand, like, s-safety issues there, or you're not really worried about someone seeing that, "Oh yeah, this person on a roof installing solar panels is AI-generated. " Like, it's not as big of a issue there. But when you compare that to, like, maybe a women's apparel brand, where, you know, you have an AI avatar wearing their leggings or a dress or whatever, and there's a certain body type associated with it, like, those are, I think, are the, the brands that have higher issue with this technology, and I think makes sense because you wanna have, like, accurate representation of how a product looks, like, on an actual person. Um, so that's where I see, I think, the biggest, kinda line in the sand at this point is, like, mainly more, like, apparel or, like, you know, skincare, like, any of these kinds of brands that are a little bit more specific to, like, how an actual person uses it. Got it. What else? What other things do we need to be thinking about, other than what you just talked about? Well, there's definitely, you know, some legality things to consider, right? There's, the FTC has-Some pretty clear guidance around the usage of AI creative in advertising. Like one, I mean, it-- people freak out about this a little bit more than I think they, they need to, 'cause the reality is the, the AI guidance is pretty much the exact same as it-- your, the guidance they give you for like a real, like a UGC creator. Like, you can't give them a fake testimony for them to say. Like, a real person saying a fake test-te- testimony is the exact same as an AI person saying a fake testimony. Um, so that's like not allowed on either fronts. Um, but there are ways to do it cleanly, one of which is kind of taking a lot of the I language. So rather than writing a script that's like, you know, "I, I, I put on this acne treatment, and my skin cleared up the next day," saying something like, you know, "This, this acne cleanser is thirty percent more effective than the leading competitor. " Like, kinda taking away the first person, but maybe it's just being presented by an avatar that looks and feels like your target audience. Got it. What about the-- What about on the platform front? Um, obviously, two of the biggest players in the online world, Google and Meta, have massive AI investments that they've made. Are they okay with the ads being made by AI? Absolutely. They have their own generative AI in the platform already. Like, they have image generators. Um, they even have, like, video generators now. I think they've... I'm might not be a hundred percent accurate with this, but I think they've even partnered with, like, Google Veo 3 or Z- or Sora 2 at this point, like, to level up their own, like, AI architecture there. Also, like, the Manus, acquisition. And yeah, I mean, they- And TikTok, TikTok's parent company, um- ByteDance. Yeah. ByteDance also has their own AI models now too, right? Yep. And they are one of the best models out there. So, every major advertising platform is kind of integrating AI, generative AI directly into their platform. So I have not heard of a single person who got creative flagged or an ad account disapproved or anything like that from using AI images or videos at this point. I think if, like this-- it's like the same rules apply, though. It's like you can't make an AI video of somebody saying, like, "You're gonna, you know, lose a hundred pounds tomorrow if you, you know, take this supplement. " Like, the s- the, the generic advertising rules still apply regardless of if it's AI or not. So that's still gonna get you in trouble. Love it. Okay, last question, and we'll get into the how side of this. Um, do you believe that this is going to become the standard here pretty soon, where most creative is going to be, either enhanced by AI or wholesale created by AI? I do. I strongly believe that's where we're headed. I think there will always be a place for, like, real content. I think it'll probably become more valuable as, like, the massive drop-- mass adoption of AI content kinda takes over. Um, so I don't think that's gonna be completely gone by any means. But I do think, you know, the platforms are trying to make it as easy as possible for some small business who's got an online store to just, like, create an account, click a few buttons, get some creatives going, and then off they go. Um, so yeah, I do very strongly think that the future is, is going to be, like, AI-generated with these static images and video creatives. Um, yeah. I love it. Okay, folks, so if you have not invested in ads because you, A, don't have access to an incredible videographer or photographer or graphic designer, I think what Caleb is, effectively alluding to is those days are rapidly declining. You have an opportunity here to really create, incredible high-quality ads that you could put across the social platforms at a very economical rate. So that is kind of gone now. So now I'm really excited about, like

let's dig in. Where do we actually start- Yeah... when it comes to AI and ad creative? Yeah, I mean, typically people get started with images first. It's much more, I would say, accessible. Uh, you know, in the last six months, we have models like Nano Banana, Nano Banana 2, I think Nano Banana... Yeah, Nano Banana 2 just came out, either last week or the week before. Um, so these models, you know, essentially are now not just image generation, but also image editing. So you can prompt your way into really any creative that you want. Um, I just built, like, a template that, allows you to just, like, see an ad in, in the wild, like, on their s- on their feed or something, screenshot it, throw it into this tool, and it will, like, clone it, like, replace your product into that image- Mm-hmm... with your brand colors and all that stuff. Like, those tools have existed for a while, but it's definitely, like, gotten a lot better, where it'll even maintain the fonts. Um, the, the text on the products has gotten extremely good at this point, too, where, like, historically you might, you know, try and throw your product into an AI image, but somewhere in the, in the process, the text on the product gets all warped or, like, replaced with different words. But that's gotten significantly better recently. Um, and what's, what's crazy about all this is, like, it's like, literally, like, in the last two weeks, it's gotten better. So this is getting better literally every week, every month there's a new model coming out. It's kinda like the space race, but for generative AI because, you know, Google makes an update to their image model, and then OpenAI makes a n- you know, an, an update to their image model, and then, you know, the ByteDance does, and all these other... Like, everyone's kinda just, like, constantly leapfrogging each other, um-So it's kind of a, it's an insane, it's an insane time. Like I, I'm in this constantly, and I'm constantly overwhelmed as well. And folks, if you don't know what Nano Banana is, and it sounds a little crazy, well, this is the model that comes with Google Gemini. So if you already have Google Gemini on your workspace account for your company, this is included for free. Nano, Nano 2 is like their Flash model. It's faster. Then they've got Nano Banana Pro, which is the model that you have to use, their more advanced Gemini Pro model for. But it's, it's pretty phenomenal. So let's kinda talk about like how we can actually use images inside of ads. Like, let's just go through some scenarios because there's a, there's a lot of things that you can do that maybe people haven't even thought processed yet. Yeah, I mean there's the traditional, I would say, like, more polished ads. Like, you go to ads library, you look at your competitors, you go to any of these other, like, ad-spying tools out there. Um, they have like libraries of ad creative. Um, you know, some of them are, like, Foreplay. co, Creative OS, Atria, Make, I'm blanking on another one, but they all kinda- On the Facebook marketplaces-- Wait, no, not marketplace. Facebook, what, Library, right? Is the- Yeah. Facebook Ads Library. Yeah. So that's a, that's a free resource- Yep... anybody can go to. They can put in, the name of the brand they wanna see the ads for. And they actually recently made a huge update to that, where you can see im-like, impression count. Like, not the actual impression count, but it'll tell you, like, high impression, low impression, so you can get a gauge of how good these ads actually are. Um, but yeah, so, like, the first method I would say is, like, using AI image generation for, like, the more polished-looking ads, ads that look like ads. There's kind of the, the basic templates out there. There's the us versus them ads. There's the product feature call-out ads. There's all these kind of, like, you know, in the d-D2C community, ads that you kinda see every brand using. Um, and Nano Banana and other models have gotten extremely good at just reproducing those with a new product or a new person or a new setting or a different font or whatever, whatever element that you're wanting to change. The second, is what I call, and, this term I'm borrowing from Barry Haot, who's kinda like the ugly ads godfather or whatever you wanna call him in the community. Um, so he, he calls these ugly ads, where, it's an ad that's meant not to look like an ad. It doesn't immediately show up in your feed and make you think, "Oh, that's an ad. " It might be something where, you know, I take my phone and I hold my product in, you know, whatever setting that feels the most natural to me on social media, and I take a picture. Or, you know, I'm using the fonts, like, from Instagram or from Snapchat or from TikTok. Um, so it looks and feels something that would just, like, blend in to the feed. This is something that I also call chameleon ads because they're meant to, like, basically take the shape of what people are used to seeing on their feeds. So that's another, that's another type of ad that you can create using AI, which, like, that is something that I've been waiting for a long time because that's my primary, use case for image ads. Like, I, I mostly make these, like, ugly ads versus, like, the polished ads 'cause I think they perform much better. But AI up until recently hasn't been able to get that right because you might prompt, you know, like, a selfie-style photo of somebody holding a product, but they have got, like, six fingers. But, like, you know, now all that's gone. Like, you can actually make image ads that look extremely realistic and native to the platform. Love it. What about people? Uh, what about generating people inside of ads? Yeah. So, generating people, super, super easy now. Um, there's a couple ways you can do it. One is just, like, expli- like describing what your character, what you want them to look like. And then, what I like to do is kind of create almost, like, a mood board of that character. This is... and this is how you get character consistency locked in pretty good, is you can create, like, eight different shots of your, your subject with, like, multiple different angles. And then you can use those shots as your references for the next phase, which is, like, putting them into an environment or having them hold your product or changing the angle or whatever you're trying to do. Um, so, like, the first step is, like, locking in that character, with multiple different angles and lighting and all that good stuff. Um, and a way that you can even do that is, like, sometimes what I'll do and w-what I've seen other people do as well is maybe they see some viral video or something, and they want... they, like, really like the vibe of the person that is in the viral video. And so all they have to do is basically take a screenshot of that person and then, kind of reverse engineer it a little bit so you can a-you can throw it into Gemini and say, like, "Write me a prompt to create a person that looks very similar to this. " So that way you're not just, like, copying someone, you know, 'cause that's illegal. You can't just, like, take someone's likeness, use it in your ads. But, you can just, like, use that person as a reference to, like, recreate them and recreate their vibe. Hmm. That's really cool. Um, so this is the g-idea of, like, kind of a reference image, right? Is that really- Yeah... what you're talking about? So you could take any picture of anything or person or place or whatever, right? And then you could upload it into, Gemini, and you could... what I'm hearing you say is ask it to describe in detail something inside of it, right? Like the setting or the person, and then you take that, and then you start experimenting with generating a new person, that looks slightly different, right? Uh, could be going from a female to a male, or it could be going from a, an older person to a younger person. I would imagine- Yeah... you experiment with that quite a bit, huh? Yeah, absolutely. And that's, that's where you get really a lot of the power of this is like you can... Like I've, I've started out with like a core person, and then I've, I've prompted, okay, I wanna have a 20-year-old version, a 25-year-old version, a 30-year-old version, a 30... Like, like five-year increments all the way up to like 65. Mm-hmm. And then all of a sudden you've got like somebody or like the, the vibe of somebody like in every kind of stage of life that now you can then use in your advertising, to kinda like hit all those different pockets of users and like be more relevant to them. Do you have any tips on how to prompt these tools to get what you want? Because obviously this is a big part of it, right? You don't just- Yeah... create the image of a person that looks like this. Any guidance on how to do that? Yeah. My opinion is that a lot of folks out there overcomplicate prompting and like prompt engineering. Um, there's way... Like I- I'm in this stuff all day every day, and I don't think I've ever written like a full prompt from scratch. Um, there's ways to just use AI to prompt for you. So what I'll usually do, if there's like a new model that comes out, like let's say, NanoBanana, NanoBanana 2 inside of Gemini just releases, right? If I go to Google and I search, okay, NanoBanana 2 prompting the, prompting reference is or prompting guideline or, like there's like a, a structure basically that, that they release on like the best way to prompt this new model. Oh, and generally the models come out with something is what you're saying? They're, they- Yeah. Okay. From the source. They'll l- they'll usually provide like an actual document of like, include this language here, include this structure, include... Here's all of the different ways that you can prompt it. Hmm. Um, like all the keywords that you can use, all that good stuff. And so what I'll ta- what I'll do is I'll take that guide and I'll throw that into ChatGPT or Claude or Gemini, whatever model you're using, and say, "You know, using this reference image, help me put together the perfect prompt for creating whatever it is that you wanna create. " Hmm. And then nine times outta 10, it will spit out like a prompt that like has all the right keywords, is formatted the right way. It might even be f- formatted in like JSON format, which is like a coding language, which, you know, you might have seen people out and about kinda talking about JSON prompting with images and how it's better. There's, you know, we could g- we could go into that. But, anyway, you can get kind of the, the end prompt pretty simply- Okay... just by talking with AI and having it help you out. What's your preferred, I use Claude all the time, but what's your preferred tool for this kind of, kind of coming up with, for lack of better words, a, a prompt template for yourself almost, where you can just substitute the stuff in and out? Yeah, I mean, historically I've used ChatGPT, but with like a, as like a custom GPT. Um, so those are I think artifacts in Claude. I also use Claude. Um, but I have most of my custom GPTs already like in the ChatGPT OpenAI, uh- Yep... ecosystem. But yeah, what I'll usually do is I'll create like a custom, GPT or like ar- artifact that has the instructions already in the context. So all I have to do is open it, and that's like where I always go- Got it... for generating new prompts. There's also like using tools like, like Airtable, like they now have, like AI agent columns. So like I put together some templates inside of Airtable where like it kinda helps me build the prompt out based off of whatever s- I select, in the table, which is also a super wa- like easy way to do Love it. Uh, for folks that are using different tools, custom GPTs in Claude are called projects, which are, effectively the same thing. And then in, in Google they call them gems. But I really like the idea of going out there and finding the instruction set and creating a custom set of instructions, and it's effectively kind of a, a resource where you could just drop in an image, tell it what you want, and then it will output presumably a better, prompt that I would imagine you could just take directly into NanoBanana Pro if that's your final source, and then just take the new image. I mean, or you don't even need the original image anymore, right? You're just ultimately- No... picking a prompt, right? Yeah.

That's crazy. Okay. So, video, let's get into video here because- Yeah... that, that is a whole nother ball of wax. Um, and I'm sure there's a direct relationship between the images and the video. But before we go there, one of the big things that I've noticed with a lot of these image generators is they don't output super high resolution images. Are you finding that is getting better? Because, for example, a lot of the text looks a little rasterized on this image stuff. Is there a way around that? Have you figured out how to solve for that? Well, I mean, NanoBanana 2, which just recently launched, they output in 4K resolution now. So that's a huge- Do you have to tell it to output in 4K? Sorry, can you repeat that? Do you have to actually tell it output it in 4K in order for it to work, you know? So if you're using it in Gemini, it's a little different than if you're using it through the, like API. So the Gemini, you're doing chatbot kind of thing. There's, there's multiple ways to access these models, and they all kind of are a little bit different. So ax- accessing NanoBanana 2 or Pro through like the Gemini chat, it's gonna give you l- slightly worse outputs than if you're going directly to the API through like their, um- Like a third-party tool? Either third-party tool or like Gemini has... Like if you just Googled, like Gemini API work, uh-The workplace. I, I can't remember the exact terminology that- AI Studio, is that what you're talking about? It is. Thank you. AI Studio, exactly. Ah, okay. So that's important, folks. AI Studio is a free tool for developers, but you don't have to be a developer. So what you're saying is you can actually generate these images directly out of AI Studio, and they're gonna be much higher resolution. Yes, exactly. Ooh, that's really important. Okay, cool. Yeah. I did not know that. And then there's also all of the third parties that, you know, have it, like Arcads, Freepik, Higgsfield, all of the, like, big kind of like AI aggregator tools out there. Mm. Um, they also have, like, the direct connection with the API, and you can get all those additional features from them. Love it. Okay, let's transition into video. Uh- Yeah... what do we need to know? So with video, the, the way it's done, I've- I would say with, like, almost every workflow right now, is you start out with your generating your reference images. So we kinda already covered getting those images created, but let's say you're wanting to make, like, an unboxing video of your product, of somebody, like, getting it in the mail, and maybe you have multiple scenes where, like, you know, you want a shot of them opening, like, picking the box up off of the, the front porch, taking it to their kitchen table, opening it up, taking the product out, like, maybe even trying it on. All those different scenes, basically, you would prompt an image of as, like, the starting reference point. And then, you would take that image once it's created, and then build out the entire video kinda like piece by piece. Um, and this is typically, I would say, how most people are doing it with models like Veo 3, or Kling, where you throw in your reference image, you say exactly what you want to have happen, if there's any dialogue, if there's any, like, sound effects, like the sound of the door opening or some crinkling of the paper or any of the things that you want to include to make it feel as, like, real as possible. And then, typically what's done is after you download all of these videos and then you go into something like CapCut or Final Cut or whatever, and then you just kinda piece it all together. Hmm. That's, that's for the last, like, six months, I would say the, the primary kinda workflow that people are, are using right now, which is like scene by scene, piece by piece. Um, there are models now that are coming out that allow multiple scenes for video. So, like, Kling 3. 0 as an example, now you can say, "Scene one, you know, girl opens front door. Scene two, she takes the box and puts it on our kitchen table. Scene three... " It'll actually put a multiple s- like, multiple sequence video together in one video for you- Hmm... which is pretty crazy. There's also, models like Sora 2, Sora 2 Pro, that is owned by OpenAI. With that you can be, like, very even just vague. Like, you can say, like, "I want a cinematic video of, you know, someone holding my product," and it will just do it. It'll just, like, wri- it'll write the script, it'll put together all the clips for you, it'll put music on it if you want it, a voiceover. It'll literally, like, put together, like, a packaged, fully legit ad for you. Um, and that's, I would say, like, Sora 2 is, like, the easiest model to work with if you're looking just to get something that's like, "I just want a final version. " The trade-off is you don't have as much control over that final output. So a lot of, like, I would say the pros are using models like Kling or Veo 3 because you can get really specific with the prompting to get, like, the exact thing you want from that. But then a model like Sora 2, you can be a little bit more vague, a little less descriptive. You can still include the reference images, so you can give it your product, a setting, you can do all that stuff, but then it'll just kinda run with it. Uh, we're gonna get into, we're gonna get into some of this. I wanna talk a little bit about AI actors a little bit because we've already, we've already talked about how you can have a reference, shot that you could have created with AI, but to actually, like, have the, the voice and the movement and the face and all that kinda stuff- Yeah... how does that actually happen? Because I... Give us a little insights. Is this completely automated now, or do you have to do some magic sauce stuff behind the scenes to make all this work? So historically, you would've had to, like, create the animated video of the actor, generate the audio in something like ElevenLabs, and then combine it all together in post. Mm. Um, that's definitely, like, super far gone. Um, we're, we're at the point now where yeah, you just say what you want, and it'll give you the avatar speaking, fully lip synced, fully animated exactly how you want, and it looks really good now. Um- Interesting. And could it be, is it, could it be your voice or somebody- Yeah... else's voice? Really? Okay. Yeah. Wow. So, Arcads as example, they have a feature right in their platform where you can just, like, click a button to replace the voice. So you can still generate the video from Veo 3 or Sora or Kling, and then you can actually connect your ElevenLabs account into their platform, and then in ElevenLabs you can clone your voice. So you can upload, like, two hours of audio of you. Like, I've done this with my voice where, like, I uploaded, like, a couple long YouTube videos, maybe a podcast episode like this, and then basically ElevenLabs, like, clones your voice, like, pretty perfectly. Um, and then you can just literally replace it onto whatever you want. So- Wow. There's lots of, of cool ways to do that now. I think the models are also adding some of this stuff, natively coming too, where you can kinda like save some of this stuff. Um... That's really helpful. Yeah. Uh, folks, for those of you that don't know it, I have another show called AI Explored, and, episode 94, which was February 24th, I had a gal named Eve Whitaker who is a Hollywood producer, and she specifically talked about the Veo 3 stuff that Ca- Caleb was talking about. Uh, not Veo 3, I'm sorry, Sora. Sora 2. And Sora is OpenAI's video model, but what's great about it is they understand the story arc, and you don't have to really have- Yep... it figured out, and it just kinda does the full arc of the story and it allows you, obviously, to place yourself into it and train up your own model of yourself and all that kind of fun stuff. Yeah. So, all right. Talk to me a little bit about, Kling, 'cause I've heard a lot about Kling, and I just recently interviewed someone earlier today specifically on Kling. Nice. But kind of explain to everybody what Kling is, and then we'll, we'll zoom in a little bit more on Veo 3, and then we'll also talk a little bit about Sora. Yeah. I'm sorry, and arc ads and all that stuff, 'cause I want people to kinda understand, like, these things and kinda, kinda what they do and what the pros and cons are. Yeah. So, Kling reset... So they're a Chinese model. Um, they recently came out with an updated version, Kling 3. 0, and, some of the biggest, like, changes here are, I would say the biggest one is that, like, scene-by-scene prompt, that now you can do. So you can have multiple... Like I, I made like a mock podcast video. Um, I gave it like a starting reference image of me and another per... I took a, like a screenshot of like the Joe Rogan Podcast studio, and then I had Gemini insert me and like another person into that studio, rebrand it so it didn't say Joe Rogan or anything, and then I uploaded that s- that like kind of studio into Kling as the reference image, and I said, "Okay, scene o- scene one, establishing shot, show both people at the table. Um, you know, the host says, you know, whatever the dialogue you want. And then scene two, have it pan to a 45-degree angle zoom-in shot of the guest. " And then, like, basically I h- I had it shot by shot switch to whoever was talking at the time with, like, what you would expect to see from like a podcast episode, and it did it perfectly. And you can't really tell, huh? That's crazy. Yeah. Is that your tool of choice generally for, for these kind of things, or do you use Veo 3 as well, or what's your thoughts on that? You know, it's one of those things where I'm co- I'm constantly bouncing around because there's always something new coming out. Yeah. Um, right now I would say I'm primarily doing Kling and Sora 2, for most things. I do leverage Veo 3 as well, but something I think that they need to work on is the, the way that their voices sound. I think they're a little, they're still a little robotic. Mm. Um, and I think Kling's, like, realism has gotten really, really good, especially for anything that's like user generated, like iPhone quality. Um, yeah. Can we take older ads and kinda make them better? Is that even an option here? You know, can we take an ad that maybe performed really well that was not AI created and somehow like enhance it? Absolutely. Yeah. I mean, there's people that will completely recreate one of their top-performing ads with AI. Um, and you- and similar, similar kinda process. Like you... If i- if, if it's a image or a video, you can upload it to something like Gemini, and Gemini's great for this because it ac- can actually, like, watch your video, and like give you feedback on it and like dissect it. But basically you can just have AI tear apart, your ad for you and like kinda put together new prompts to recreate it. And then you can like just swap out things. Um, like Kling and, I'm blanking on the other model, but both of them have, um... Sora. Like... So, so they both, they both have this ability to swap the main character. Okay. So like you could have, I could have an ad where I was the main subject, and I could throw it into the tool and say, "Swap it out with this new reference image," and maybe it's like turning me into a, a female or turning me into like an older version of myself or whatever. Um, and then it'll literally, like, it'll put the new person and it will follow all of your movement, it'll follow your mouth, like everything will be exactly locked in. You, you might see a lot of content like on Instagram or, or TikTok of people kinda showcasing this, where they'll be like sitting in front of their camera and then all of a sudden they're like a celebrity or they're like Mario or whatever. Um, that's, that's kind of the, the technology that's being used for that. Very interesting. Um, what about, what about like... What about sound effects and, you know, all the other things that you can potentially do to make, you know, these things perform a little bit better? Yeah. So a lot of models are also getting really, really good at incorporating sound effects directly, like are, like printed onto the video that they give you. So Sora 2, super good at that. Like it will... Like I had... I was like, "Create a, a fight sequence of me, fighting like a monster or whatever," and then it's like, it adds the like laser sound effects, the punching sound effects, the like music, the dramatic scene. Like, can put something together that looks like it was like an Avatar film. Wow. Um, and kinda use, and just do it. You know, you don't really have to even prompt it. There's also ways where you can prompt exactly what you want with some of these models, like, you know, insert this sound at this time and all that good stuff. But yeah, and then, you know, there's also the third party kinda like other way to do it, which you can go to something like ElevenLabs, and you can, you can prompt any sound, any sound effect that you might want now. Um, where it's like if you want like a bicycle bell, or if you want a, you know, someone ripping a card- cardboard box or whatever it is. Like, you can just literally put that into something like ElevenLabs, and then it just generates it for you. What about, formats, like 16-by-9 versus 9-by-16, you know, vertical- Yeah... or app kinda stuff? Does it with that kind of stuff? Yeah, super good. And this is actually one of the biggest, I would say, slept-on use cases for advertisers, specifically for vi- or for images. It does video, too, but images it does this really, really well. Um, like I ju- I just did this in like this air table template that I put together where, as like the last step in my whole workflow, I had to took, take the AI ad that I put together, and then I have a column for each of the aspect ratios that I want, so like a 9-by-16, a 1-by-1, and a 4-by-3. And then you just like click a button, and it just automatically takes all the like... It's able to basically like vectorize your entire ad and then just like put it square without it like squi- squishing things or stretching things. Like, it's not just like taking the image and, like shrinking it. It's like actually rebuilding the image- Will it add context too? If it's a 9-by-16 but it was filmed, landscape, it'll add more of the real context- Yeah... right? Yep, yep. Yeah, exactly. So there's ways to do that, yeah, for image and for video, where you can kinda like take an asset you already have, throw it into the model, have it create the other dimensions that you need, and then yeah, to your point, it'll even like fill in, where there's, there's gaps. For folks that have been paying attention, you don't have to do this with ads. You could also do this with organic content. Oh, yeah. 'Cause I'm guessing, Caleb, you are creating like Reels and Shorts and all that kinda stuff and TikToks, uh- Yeah... probably to experiment with this stuff before you turn some of it into ads? Oh, yeah, totally. Um, I mean, a lot of the content I make is showing how to use these tools, so I'm using it in my organic content all the time. But what's been a really interesting subject, and you could ha- You should probably find somebody to interview as like a whole separate episode for this because I think this is really, really interesting, but there's AI channels now that are really popping up. Um, like there's one, I think it's called Chloe- It's like, it's like faceless video, but it's all AI is what you're saying, right? It's like beyond faceless video. It's like pers- AI personalities. Mm. So like, influencers that are not real, but like every video is that inf- that fake person. Like, there's one Chloe versus History, and like every, all, every Reel is this girl. She's not real. She's AI. But it's like her in different like scenes throughout history, like the Revolutionary War or Paris during the Black Plague or all these things, and it's just like... Or like on the Titanic when it's sinking, and it's like a minute-long video, and it actually is like really good and really entertaining. Like, it's like not just AI slop. So I think it's really interesting to see on the organic side the way that people are using these tools to also just like get mass engagement. Like, her videos have like 7 million views because people are entertained by it. That's crazy. Caleb, we have just scratched the surface of AI creative, for ads. I know that there's gonna be a lot of people that wanna connect with you. Uh, where's your preferred socials if they wanna connect with you on the socials, and if they're potentially interested in, in reaching out to you 'cause they wanna be part of your school or anything else you got going on, where do you wanna send them? Yeah. So, you know, online my like persona is like the MrPaidSocial thing. So all of my channels are just @MrPaidSocial. Some of, I think on Instagram and TikTok it's like Mr. PaidSocial. Mm. And then on YouTube I'm one word, MrPaidSocial. I'm also on LinkedIn. Um, and then yeah, I've got my private community on school, the AI Ad Alchemists. And, yeah, that's about it. Caleb, thank you so much for answering all my questions and sharing, your wisdom and insights with us today. It was super insightful. Absolutely. Thank you for having me. It was super fun.