How to make high converting ads with AI (Raw Build)

In this video, we're gonna be building a UGC creative workflow using Pletor AI. This is a really cool tool that we've been using inside of our business, and I wanna build a workflow with you totally live so you can see how to use it, how we're using it inside our business, and how you can really increase your creative production without necessarily needing to hire a ton of team or do anything too crazy. So let's get right into it. What the purpose is of this agent that we're building. Is for UGC AI video creatives. So this is going to give us an end result where we can put UGC AI videos up and running in just a few minutes, and I'll show you an example of an end result here right now. So this is inside a brand new ad account using a new-ish offer that we've been testing. Overall, we've been getting some pretty good results and we're really profitable over here. And this is the end result of a UGC AI video that's been doing well for us.

You're about to send her a heart.

So it's fairly simple. We're going to use Pletor to build the whole thing out. Let's get right into it. So what I would recommend is if you're feeling stuck and if you're not too familiar with the software, get familiar with it. But how it basically works is you've got these nodes that you can input and then connect them to outputs as well. I have an image add one that I'll show you real quick, and you can see how I've structured this one out is we basically have these inputs where I put in the hook, the visual direction I want, offer details, a target audience. I have a very specific prompt that outputs something here for me, and then that feeds it into 4 different AI image generation nodes. So I'm using Nano Banana, Flux, GPT, and then Seedream. And then I have an image upscaler here which will then give me the finished creatives. Obviously not all of them are great, Um, but there's one or two in here that I like and we end up using them and they do really well. We're gonna be building out the same thing here for the UGC ads, just a bit different. And what I really recommend is to use something like Claude or ChatGPT to help build everything for you, or at least give you the foundations, cuz it can be a lot to like come to this page and have a totally blank canvas and not sure what to do. Um, but let's get right into it. So first things first, we're gonna add in a couple of input nodes. So for these UGC ads, I like to be really clear with like the actor demographics, 'cause sometimes it's nice to test different looking people. So actor demographics will be one input. We're gonna do another input, which is gonna be the offer overview. So the AI kind of understands who we're going after based on the offer. We're going to do another one, which is going to be the hook. Put that here. So that'll be like the first few things that the AI actor is going to say. And then I do like to put in a desired video length as well. It's just nice to feed that to the next step here when we're inputting everything. So those are all of our inputs. That's everything that we're essentially going to be feeding to the AI in the next step here, and that's those inputs. Now, the next step, the next step here is to take all this information that we're going to be feeding the AI and to actually feed it to AI in the first place. So we're going to make a new node and it's going to be a generate text one and we're going to configure this here. So from here you can choose a ton of different models as far as everything, depending on how much you want to spend credit-wise and what kind of result you want, you can adjust these things. I do like using something like CloudSonnet. Gemini 3.1 Pro and Gemini 3 Flash is pretty decent, but CloudSonnet is, I think, is really, really good. So we're going to use that. And then I wrote a prompt here that we can use, and I'll read it out here real quick. So basically says you're a direct response video ad script writer using all the inputs, write a full UGC script that's exactly input 4 long. UGC actor input 1, offer input 2. So these are all the inputs which I'll change. Break the script into 6 to 8 scenes. Each scene must be 8 to 10 seconds long. For each scene, output scene number and the timestamp, right? Etc., etc., etc. So that's great. I'm going to keep that there for now, but what we have to do now is actually connect all these nodes into this one. Now we've got all that information, and then we can kind of do all this here too. So we can say video ad script that is exactly desired video length long. UGC actor would be UGC actor demographics. Offer will be offer overview. Hook, which is going to be the hook line, and that's it. And then it should be able to pull them for everything else. So that's everything there that will generate us basically the prompts to then go and make everything else that we're getting. So what I'm going to do from here is do a scene splitter output. So let me show you how this works. Actually, I'm going to tweak things a little bit just to make sure that this works. So I'm actually going to go back here and I gave it some more rules. I think that's important. I made it a little bit more clear as far as like the scene stuff and then, and then more details here. So that should be a little bit more dialed in. You can, you can pause and kind of read through the prompt that I gave here, but it's essentially the same as the first one. I just tweaked it a bit. Okay, so after that point, now I actually want to generate the influencer persona. I want to tell AI what the person actually looks like so it looks good and it looks consistent throughout. So what we're going to do is we're going to do another generate text here, and we're going to put that up here because this here is basically generate scenes, and this one here is create Influencer. We could just call it influencer persona, actually. Persona. And then I'm gonna do the same thing. We're gonna go Claude Sonnet, and we're gonna do this. This one basically says, hey, you're a visual character designer. We're gonna use these for all video scenes, actor demographics. Make sure I put in UGC scene splitter, yada yada yada. And it basically has all this, right? So that we want to connect the UGC actor demographics. And then here we're going to use the UGC generator, and I'm blanking on what it is called. Higgsfield. Higgsfield makes really, really good AI models. So we're going to use that and it's going to make an AI image based on this influencer persona. So that's the next step. Okay, so I tweaked things here just a little bit as I was reviewing it. So there's a couple things here. So we've got all of our inputs going into this main image or this main text prompt, which will generate the scenes. Then we've got up here the influencer persona being made. So it makes the influencer persona, then it actually— I have a prompt in here that will make a really good reference image. So you can see all this stuff here, this prompt. I'm basically saying, hey, you're a prompt engineer using the influencer persona that we just generated. Make me a photorealistic reference photo of this person, right? And that's from here. And then from there we put it to Higgsfield Soul and that will then make the AI image and that's going to be based off the reference image prompt. Then the next step is to actually generate each of the scenes. So what we've got here is I've got this text prompt again. And I'm saying, hey, you're a prompt engineer using the influencer persona and the scene breakdown. Write a video generation prompt for scene 1 only. Right. And so that'll be scene 2, or that'll be for scene 1. And then we're just going to duplicate this and do the exact same thing, but it will be for scene 2. So we've got scene 1, we've got scene 2, and we'll change this to say scene 2. Let me make sure I don't have scene 1 anywhere else. Great. And then we're going to do the exact same thing for— how many scenes are we getting?

7 scenes.

So we're basically going to do the same thing, uh, 7 times. So scene 3, 4, 5, 6, and 7. Right, and then we can just go in here and edit all these. I'll change them. Okay, so I've got everything pulled out for each scene. We're pulling the influencer persona, etc., etc. Now what we're going to do before we link everything else up is I'm going to put a video generation thing here, and we're just going to test it for the first scene actually. Just so we don't spend a ton of credits to see what's exactly working. Because if the first scene works, I know that the rest of them should work pretty well. So I'll just do the AI video for scene 1, and let me get a prompt for that. Okay, great. And you can see here for the AI videos, you can't really feed it like a very specific prompt in here. So to make sure that it actually does the right thing, we're gonna put another generate text here. in between each scene. Like I said, for now I'm just going to do this for the first scene, just so we don't spend too many credits and something is broken. And then basically what I'm saying is, hey, you're a prompt engineer for Kling, you'll receive a scene description, etc., etc., please make this thing. And what I'm going to do is I'm going to also put that influencer persona here, and we will connect this to here. So that should do it. We've got the generate text, and then finally we've got the AI video. We're going to connect that prompt that we just did. We're going to connect the AI image that was generated, which is going to be our prompt. And let me see, should generate it. We'll do 10 seconds here. I was thinking about doing Sora because you could do a 12-second video for Sora. Um, I may do that because Sora just pops out some really good videos, so I may actually tweak some of that. Okay, so I think I've got everything hooked up how I want it to. Um, I will— I'm going to generate this first one and see exactly what we get, but I'm gonna walk through it here real quick too. And I just realized I didn't even give it any, any inputs, so that does not help. I wish you could pause it, but you can't. So I'm going to just wait. Okay, so it looks like it just picked some random stuff to make because I didn't give it any inputs, unfortunately, because I was too excited and clicked the button. But it's made some general stuff— 32-year-old female, etc., etc. And you can see, so it's made the prompts for everything. It's going through now and it's made actually all of the scenes matching with the influencer persona that we have. So there's that picture that we got, and now it looks like it's generating this video. Let's see exactly what it does. Obviously this would be a little bit more dialed in if I put in my prompt, but I, I did not do that, but that's okay. Um, I don't know how much time I can spend just like moving these around so it looks perfect. But let's see exactly what we get. And while this video is generating, let's actually check to make sure that it generated the rest of the scenes correctly. Um, so let's see, I'm gonna zoom in here. All right, so it's basically got details about the person. Plus eight to ten seconds. What's the difference between each one of these? The only thing is, I don't know if this generated the exact script that I want to use, but it seems like all these are almost too similar. Because this is— let's see the scenes. Young mom subscription box, mom reached out of snack time. Once you have the details, I'll make your script. Okay, yeah, see, so it actually may not have worked very well because I didn't even give it the details. So I'm gonna wait for this one to finish and then we'll see. Okay, so I've got everything hooked up and I've got some prompts in here. We're going to go ahead and fire this off and see how it does. Okay, great. So it started to generate everything. Obviously, I'm not going to get all my scenes as a video. However, I am getting every scene here now. So this is where we can kind of check everything. It's showing each scene, scene 1, scene 2, scene 3, scene 4, all the way through 7, which is great. And what we want to make sure of is technically in each one of these nodes, The scene should be slightly different. And it seems like that's the case. And if all these look good, I'll probably hook up the rest of the video nodes and then generate that video and see exactly how it does. I'm not sure why this one hasn't generated yet. That's because it's waiting for that image to be generated. So these workflows are really cool. You can do a lot of really cool stuff with them, but we'll see what we get. So we got the image made, and now it's going to start generating that video. If this first video is good, I'll get the rest of them to generate, but what we'll end up doing once all of them get generated is you actually have a final node over here where we can basically merge all the videos together. So we'll basically just connect all of them here, merge them, and then you technically could do like an upscale video tool as well if you wanted to. Because you can see here Sora was generating this in 720p, so if I wanted to upscale it a little bit I could, um, but that would be great. All the videos would be generated, and then I could basically have, um, these set over here somewhere to feed the rest of them, and we would be pretty much good to go. So really curious to see how this video turns out. Just gotta wait for that now. All right guys, so we got the first video generated. Sora's taking too long, so I just plugged Kling in. It's not exactly how I want it to look, and she speaks like gibberish at the end, which isn't helpful at all. But the video itself looks pretty good. But I want to simplify this a bit because I think too many prompts in between the final output is just not good. So what we're going to do is we're basically going to have 2 prompts here. So we're just going to tweak this a bit. I'm here in this generate scenes. I'm going to clear this prompt and add this one here. So I'm basically telling it, hey, I need you to get me, or I'm going to give you a hook, the actor, the offer, the hook, target length, influencer persona, write 7 scenes, et cetera, et cetera. And then give the script as well. So that will generate that, which is great. So that's a new, a new one there. And then for the scenes, Um, takes one scene from the output above, converts it into the clean prompt, and then duplicate it per time per scene. So we're going to go and put this in here, and that's pretty much it. So this is going to simplify it a lot more. So I'm actually going to delete that node. We're going to pull this up. I'm going to delete these for now just to clean this up a little bit. So, and here's what we've got. So we've got This one, which is going to bring that, and then this one, which is going to connect here. So is this connected to what I want? I believe so. Make sure scene 1—

actually, I don't know if it is. Is it? It should be.

I keep clicking the video, not the actual prompt. There we go. Okay, cool, great. That should, that should do it a little bit better this time. Let's go ahead and test run it again. And you can see why it's important just to test it once, like if you have a big workflow, because eventually what we want to do is we want to have it make AI videos, but what if we had 3 different hooks, right? So then that way we need to generate all the scenes to put together like a 45-second to a minute-long video. But then I would also want different hooks to stitch on the beginning of it. That would be really cool. But for now, I'm just wanting to do one at a time, or else you're going to burn through credits so fast and it's just not going to be worth it. So we'll see here. We're getting, we're getting this AI image. Hopefully that plugs it in correctly here, and then we should get a good output. Sora was just taking way too long. I may plug it back in in the future, but right now it's just taking forever. So, and Kling looks pretty good as well as far as like what the person, what the AI actor actually looks like. So we'll wait a little bit longer. We're waiting for this picture to be generated, and then this AI video should kick off. Okay, so I got the video to generate. She's speaking this time, which is great. But the actual video itself is not that great because she keeps getting closer and closer and closer to the camera. And like exactly what she's saying is a little off. So I think I'm going to simplify this even more. So what I did is I asked Claude to give me a simpler prompt. And what I'm basically going to do is instead of all of this, I'm basically just going to have it go because I can always get this back. So I'll delete these. I think having too many prompts is just confusing it. Basically, this reference image is probably the most important thing to keep the similarity the same between each scene. And this one looked great. It has each scene, it has the timestamp output like this. This looks good. So basically what we're going to do is we're going to send this to Another generate text here. Connect these two, and I'm going to use Claude again. And then I've got this prompt here, but I'm going to check it again. Prompt engineer, you'll receive seven scene breakdown. Your job is to only write one video generation for scene one that produces both scenes spoken together. Yada yada yada.

Okay, great.

So that should give. Us everything for scene 1. So I'll just say scene 1 video prompt, and then that will plug in to this video now, and that should be a lot better. And then I'll just have to duplicate that for every single scene and say only generate scene 2. Let's test run it again because I like what it's generating here in the scenes, and I like the model it's generating. I think that's great. But just this final output video is close, but it's not quite there. This just shows you have to test it a few times. Usually within the first few times you'll figure it out. But the good thing is for you guys, you could just kind of steal the work that I've done, but you can at least see how to build them, which I think is important. But this first piece is, is great. It's generating a good scene. It's generating a good script for each one. It's showing a start and an end to each scene. Let's see. Don't understand. I could probably make the hook and whatnot better, but let's see if it does a better job here. Okay, so I think we finally got a good prompt and a good output. I'm probably going to change and kind of upgrade all these ones, but You can see here it generated it. The script here is new moms everywhere skipping everything for this. And honestly, after trying it myself, I totally get why. It's a game changer. Seriously, look at the look at the output here. It's actually not that bad.

New moms everywhere are literally skipping everything for this. And honestly, after trying it myself, I totally get why. Like it's just a Game changer. Seriously.

So that's the first one. I think it looks pretty good. If we upscaled it, maybe if we were a little bit more specific on some of the details of the influencer persona, we could do that. But overall, I think this looks pretty solid. Um, let me refine this and do a little bit more detail on all of this, and I think we'll get a better output as well. So let's do that. Okay, so it took a little bit of testing, but I'm definitely getting to where I want everything to be. And Sora is generating a little bit faster. So I did one with Grok, wasn't perfect. I did one with Sea Dance, it was pretty decent but didn't nail the accent. But I'll still share with you this one. Sora 2 definitely did a good job, but I accidentally just had it regenerate. But let me show you these ones. Okay, so here's this one on VO3. Obviously the aspect ratio would need to be changed a little bit, but it's pretty good.

This AI tool could Find winning products, eh? And what if you could make an extra $5,000 a month just 30 minutes a day today?

So it's okay, um, accent-wise I'm probably throwing it off.

Today everyone, what if this AI tool could find winning products, eh? And what if you could make an extra $5,000 a month just 30 minutes a day straight from your phone?

So not bad, right? Obviously we then— and Sora 2 generates a really, really strong version of this. But essentially what we could then do is we've got that one scene. We would just do the same thing and do scene 2, 3, 4, 5, and 6 and then stitch them all together. So this is a really cool workflow. This is something that we're going to be consistently building out internally. And we also have like a really simple one too here where you can keep it a lot more straightforward, where you can do something like this. You give an avatar description, the exact script you want, any brand context. You'll then get a creative brief, creative influencer persona. You get the image still, it kind of crops it to the 9 by 16 format. Then you give it the facecam prompt and then you get this.

Mate, I just worked 12 hours for the 3rd day straight this week, like busting my ass in the heat. I still can't afford to take the missus out for dinner without checking the bank balance first. That sounds like you can.

This is a great like hook right now. This could be a hook for the main video and then maybe I generate the main video somewhere else. Or within the same workflow, right? So that's everything there. I hope you guys enjoyed this video. If you did, like and subscribe. Definitely going to be making more content like this, but I just wanted to walk you through a very raw and real way to creating this content here inside of Pleitor.