Loading summary
Transcript83 lines
- [00:01]
A
On today's show we have Nate Herc, who is an N8 ninja and he's got us an agent that is going to take any image and turn it into a remarkable image based ad or video based ad for your business. And he's giving us a free template so anybody can go build that agent for themselves. Let's get to today's show. Foreign. Hey, Nate, welcome to Marking against the Grain. We're so excited to have you here today. We're going to talk some advanced AI workflows and you've got a really cool workflow that's going to take an image and turn that into an ad and kind of show everybody what's possible with these AI workflow tools and these new image and video models. I'm really excited. So thank you for coming on the show today.
- [00:57]
B
Shout out to you, Nate, for absolutely crushing YouTube. How long ago did you start on YouTube? Like a year ago?
- [01:04]
C
Yeah, pretty much a year ago to the date it was September 19th was my first video.
- [01:08]
B
And you are one year, 389,000 subscribers. So, man, you are absolutely killing it on there. And so highly recommend your YouTube channel as well. AI automation, you have like a ton of cool stuff and it's cool that you're going to get to go through it live for our listeners and our watchers on YouTube. So I'm looking forward to learning this.
- [01:30]
C
Absolutely. Yeah. I appreciate the kind words. So, yeah, I'm ready to just jump into it if you guys are.
- [01:34]
A
Yeah, let's do it, man. For everybody watching, Nate's gonna take us through a workflow. We're gonna talk about other workflows. We're gonna build something in this episode. So it's really about how can you build agentic workflows for your business. And so, Nate, let's kick us off, man.
- [01:47]
C
All right, so what we're looking at here in Nitin, this is my Mr. Photoshop agent. And I called it Photoshop Agent because the full video I did on YouTube, I did this when the Nano Banana model dropped, which was Google's model to turn text into images and also kind of edit existing images. And that's what we have down here. We have these two tools that are for image generation. But for today, I added on a special extra tool which uses Google's VO3 model that can turn text to video or images to video. So you can see, basically this is the agent talk to it through Telegram and it can do image actions as well as a video action. So what I'm gonna do just to Start us off here is I'm gonna open up Telegram in just a sec and I'm going to just drop in an image and then we will be able to have the agent sort of like turn it into an ad for us.
- [02:34]
A
The screen looks scary, Nate. It looks scary.
- [02:37]
C
So N8N is really no different from your Zapier, your make, your Alteryx, Power bi, any of these workflow automation platforms that have been around for a long, long time. Because automation has been around for a long, long time. But now because the fancy word AI has been slapped in front of everything, it feels like something new. So the reason that I chose Ennadent is because it's a super visual builder and you'll see what I mean by that. When we start running this demo, you'll see the nodes turning green, you'll see the agent thinking it's just really cool to see what's happening because what's happening in the backend is there's a bunch of code and JSON going on that's actually making this stuff happen. But because now we're in a no code drag and drop interface, it really lowers the barrier to entry for people to get in here. So just for your guys context, I have no coding background. I went to school for business analytics and marketing and I've been able to just spin up some really cool stuff that would have taken a traditional software engineer weeks and weeks, but it can take me a few hours in a day, which is really cool. But the possibilities of NN are endless because you have all these different actions. You can build agents that can call on different tools, you can have these tools be different workflows, you can have the tools be different agents. So I know that sounds like a lot, but the point I'm trying to make here is you can literally do anything and automate anything in edin. So what we have here, like I said, is this is the main agent. So this kind of controls everything and we're able to talk to this agent through Telegram. So I could pull up Telegram on my phone for the sake of the demo. I'm just going to have it right here on the screen so you guys can see. And what I'm going to do is just kind of walk through what I'm doing and show you guys kind of step by step what the agent's doing as well. So because we're in test mode, I just have to click this button. So the agent's now listening to us and when I send over this image, so we've got this Image of me holding a JBL speaker. What's going on right here is you can see the visual element here. The agent just received that image, it uploaded it to Google Drive, and now the agent is thinking about what to do. So you're gonna see right here in my Telegram window, it's gonna come back and say, awesome, what would you like me to name that photo in your Google Drive? So now I can just say, okay, cool, let's call that Speaker. I'll call it Speaker Image. So now once again, the agent has received our message, it's thinking about it and it's going to go ahead and use this tool down here in Google Drive called Change Name. So it takes my natural language request, it understands context of what's going on and it uses the tool that best suits the use case. So it knew it needed to change name. And if I now go to the actual folder in here that it used and we refresh this, you can see that it just added a picture right here called Speaker Image. And this is the one that we just saw me upload into our Telegram. So it has access to this folder and it was able to change the name of this image just that quickly through my natural language request. So now that we have that image uploaded, I'm basically just going to ask it to make it into a professional sort of like studio looking image and then we can take that image and turn that into a video. So let's say we want to make some sort of, you know, like Instagram ad where we want to, you know, like targeted for summer, having summer pool parties or something like that. So I could say, please turn that speaker image into a professional studio. We'll just do image where the speaker is next to a pool for summer. Okay, we'll see the context that we gave it there. And now the agent once again is thinking with its brain over here, GPT5, it's going to try to figure out, okay, you know, I have this picture, I have this request from the user, which tool do I take action with? And what it should be doing is it needs to understand what file to actually edit and then it's going to use this Edit image tool. So one question that I get a lot is when you're building these agents, which chat model do you hook up? And basically just the brain. So you know, you've got like your anthropic Claude models, your OpenAI GPT models, Gemini, you've got all these different models and they all just have like different strengths and weaknesses. And right now I'm going with GPT5, which is actually on the slower side. But I found the performance to be better. Really good. And later I'll talk a little bit more about, like, what goes into prompting images and videos and stuff like that. That's where all the magic lies is in your prompts. So you can see the agent finished up. And now back in Telegram, we have this image where we have the speaker and it is by the pool. And the agent said, okay, I called this image speaker by Pool Summer Studio. So we have that picture done and you can see if you look at the actual JBL image, it has consistency with, you know, the wording, with the buttons, with even the handle thing that I use to carry it around. The value here is this model, Nana Banana. It can actually do good with like, text and it can have consistent characters because typically if you would have used an image generator, it probably would have messed up the wording or it just wouldn't look like the exact same speaker. It would just look like a very generic speaker.
- [07:18]
A
Yeah, I think it's really important to tell everybody that up until Nano Banana came out, which is what, like 2ish months ago now, like, this would have been really hard and not really been possible to do this use case. And this is why the changing dynamics of AI can be tricky, because you gotta know what the core models are capable of so that, you know, you can go and build a workflow like this. And so, you know, part of the magic that you've done here is you understood what NANOBANANA and GPT5 were capable of. And now you've been able to take that and put that in N8N to actually scale that and get a agent that you can do repetitive tasks with.
- [07:55]
C
Exactly, exactly. It's all about figuring out your process and what you can automate and what you can't. Because when I dropped this video, a lot of people were like, okay, that's cool. But you could literally just go to nanobanana in your browser if you want to drop in an image and have it be edited. And that's 100% true.
- [08:10]
B
True.
- [08:11]
C
The way that you really unlock the value of like a scalable system is, is when you have everything running kind of on the back end, where you don't even have to think about it. So you could have a system where every morning it's going out and it's doing research on things that are going on in your industry and then it's coming back with a bunch of research. It's looking at documents from your most recent meetings or your most recent news in your business. And then it's maybe creating an image for LinkedIn post or like a carousel for a LinkedIn post. And now all this stuff is just happening on the back end without you having to think about it. But like I said for today, I wanted to show this agent demo because I think that it's a lot more cool to see the agents thinking and calling on us. Different tools. So now that we have this speaker image, what I want to do is turn this image into a video. So we're once again going to have the agent start to listen to us, and I'm going to ask it to turn that speaker by pool summer studio image into a video VFX add. And actually we're just gonna leave it at that. And we're gonna see what it's able to do with just a very minimal.
- [09:15]
A
Prompt, super low context, like barely telling it anything. Yeah, I like to do that a lot of the times to just be like, all right, well, what kind of context does it need? And I'll know that by giving it, like, basically no context.
- [09:26]
C
Exactly. Yeah. So it should be interesting. Cause it's gonna get very, very creative here. One of the cool things about VO3, which we can look at later, is you can tell it to have dialogue. You know, you can tell it in the video. I want the person to say this, and the person will say it. Or you could say, you know exactly, very explicitly what you want to happen in the video. And we're basically just going to let ChatGPT5 figure out what it thinks. A nice VFX ad.
- [09:49]
A
I think that's a really good thing. While it's working here, what's really happening is this prompt is getting passed to GPT5, and GPT5 is looking at this photo and saying, hmm, what kind of ad do? I think it would be good. But it doesn't have the context of where this ad is running, who. Who this ad is for exactly some of the basic things you would do when making an ad. So it's basically gonna fill that in because you have this agent set up in a way. It's not gonna come back and ask you a ton of questions. It might come back and ask you one, but it's going to kind of give it its best broad guess.
- [10:18]
C
Exactly. Yeah. So it's gonna be interesting. We'll see what it comes up with. Right now it's hitting this, this tool, which will take like a minute to actually make that video. But yeah, just one thing. I thought I'd real quick show you guys is the system prompts of this agent. So this is the agent and this is basically its instructions. And this is very, very minimal. I'm saying, you know, you're a personal assistant agent, you have tools to help the user with their request. Here are the tools you have. And I don't want to get too technical here, but I'm doing very minimal prompting for visual ad. I'm basically just saying, you know, always structure your prompts to receive hyper realistic results. You can talk about the camera, the lighting, the composition, all this kind of stuff, but it's very, very minimal. So the GPT5 model has full autonomy here to create whatever it wants.
- [11:03]
B
Look, Nate has given you his five AI workflows he uses to build multiple ad campaigns in minutes. You can create, edit and deploy professional ads across Facebook, Instagram, LinkedIn, TikTok, you name it. From Photoshop level images to high quality videos, you can build any ad with Nate's workflows. He's given it to you for free. Scan the QR code or click the link in the description. Now let's get back to the show.
- [11:28]
C
And you can see here what it did was it searched the AI images folder, which is a different folder that I have in my Google Drive right here where we can see we have the speaker by Pool summer studio. So this is the one that it just created for us. And we basically just have, you know, separate environments for different types of pictures. And it's important because the agent needs to understand which picture to actually send to VO3 or NANOBANANA. Yeah, so that's why it's able to.
- [11:52]
A
And I think this is what's really important while this is finishing up Nate, that for people who are unfamiliar with these AI workflow tools, what you've built is a custom agent here. And the couple of interesting things is it has certain places where you're storing files. In this case Google Drive, you have certain brains, GPT5 or Claude Sonet. Right. And it's going to give you output and you are prompting it through Telegram, but you could prompt it through a whole host of different systems, whatever your system of choice to like send that information. But once you get it up and running, you shouldn't need to really be in here that much. Right. Instead you're going to work on whatever that system where you're doing the prompting and the triggering from. And again in this case it's Telegram. But you don't have to be doing the step by step in eight and all the time, once you've got that workflow built out, right?
- [12:43]
C
Yes, absolutely. 100% right. And yeah, that's exactly what I was hitting on a bit when I was saying, like, the best scalable systems with the highest ROI are the ones where you don't even touch them. You don't even know they're running.
- [12:54]
A
Exactly.
- [12:54]
C
They're just always going on in the back end. You know, like a new form submission is submitted on your website. Okay, well, we are going to have an agent in the back end, research them, reach out to them, give the internal salesperson a brief, all this kind of stuff, and that grows the business. And then in turn, the system just gets used more and more, which is where it gets really cool. Yeah, great, great point there. So it comes back and you can see that there's some weird text going on. And that's just because we didn't tell it what to do. And we gave it very minimal prompting, but you could hear it. It added music automatically. It kind of zooms in on the speaker and it has context to know it is a speaker. So it says like, you know, turn up, summer, all day, battery waterproof, all this kind of stuff. So given the fact that we gave.
- [13:33]
A
It literally no guidance.
- [13:34]
C
15 words. Yeah, exactly. Given the fact that it had nothing to work with, that's not too bad for a first iteration. What I thought maybe I could show real quick would be one more quick example where let's say we have a different image. We'll use like this cologne image. So I have this cologne image in my folder and we'll do the same thing, but this time we'll have it actually, you know, we'll kind of prompt it on making an ad that's tailored towards a specific type of person. So first of all, I'm going to say please take the Cologne image and turn it into a professional looking photo, let's say on a bar, like where you'd get a drink. So we're completely context switching here. It was just talking about the speaker for multiple messages in a row and now it's going to have to basically reset and try to figure out, I have to go find that Cologne image.
- [14:28]
A
Which is in your Google Drive for everybody following. It's searching Google Drive, right?
- [14:32]
C
Exactly. Yep. And it searched both folders and now it's able to see, okay, I found that image. I'm gonna send it over here to the Edit image tool to turn that into a better one. So we'll wait for this to come back. And there we go. We can see that now it was able to hit the Edit image tool and once again it's hitting all of these different, you know, servers in the backend. So these ones are hitting the Nano Banana models, this one's hitting the VO3 model and this is where you can customize. You know, maybe you want to use ChatGPT's image generation or maybe you want to use the new SOAR model. You can just come in here and basically change what's going on in those tools.
- [15:04]
A
So we've got it thinking here right now and basically what we're doing now is you're having it create a new kind of professional image and then we're going to give it some more advanced ad instructions to show that with the right context you can actually generate a good add in product, right?
- [15:20]
C
Yep, absolutely. And so here's the image we got back. You can see the words are the exact same. It says Nautica Voyage. It's the same as pretty much the one that I uploaded here where I was just holding that bottle of cologne. And you can also tell in this image it's pretty low quality.
- [15:33]
A
Yeah.
- [15:34]
C
But we were able to get this one back that, you know, it looks professional. It pretty much is the exact same shape and the words are consistent.
- [15:40]
B
That's cool.
- [15:41]
A
It's pretty sick, I gotta say.
- [15:42]
B
Right.
- [15:42]
A
Kieran just gave a. Wow.
- [15:44]
C
It is pretty cool.
- [15:46]
B
How much is the model, the differentiator between the results or the prompt then like can you just speak to how valuable it is to iterate on the prompt and then. And how much value have you seen increase as the models have continually got better?
- [15:58]
C
Yeah, so the prompts are a big deal. But the best prompt in, you know, an AI image generator model from a year ago would not beat. Yeah, the worst prompts from like nanobanana. Nanobanana is just so, so good under the hood.
- [16:10]
B
Infinitely better.
- [16:11]
C
Yeah, yeah, infinitely better. And also they're getting cheaper and cheaper to run too. It used to be like 25 cents an image and it's getting more and more accessible, which is pretty cool as well. So let's see what we can do here with turning this into a video. I'm going to actually whisper flow this so I can get a little more context. We're going to say please take that cologne on upscale bar image and turn it into a video. It should be a professional studio grade video that we can use for advertisements and we want to target this towards, let's say men in their 40s. And the feel of the video ad should be Professional and minimal. Okay, we'll see what it's gonna do with this. So we gave it a little more context, we gave it a little more information as to who they should be sort of appealing to. And one thing I wanted to touch on here is remember how I showed you guys the system prompt we're in here? We have just like the most basic things about prompting for image or videos. The reason I did it this way is because I built this so it could be a very general template so I could give it out to people and they could play with it.
- [17:17]
A
Yep.
- [17:17]
C
If you wanted to take on this template, you could load in that system prompt with, hey, here's my business, here's the products we make, here's the people we appeal to, our target avatar, here's our brand, our mission, our vision, all this kind of stuff. And the more context and more subject matter expertise that you give it related to you and how you want your creatives to turn out, the better and better it's going to be. And you can also start to give it good examples, bad examples. So that's where these things get very powerful, is with the more context that you give them about you.
- [17:46]
B
One thing I do, Nate, is I create onboarding docs for tools. And so let's imagine you're recreating your flow and you are kind of creating the agent who does the V3 videos. There's still some amount of knowledge of how to extract like good videos from VO3 you need. But if you're just trying to get something up and running one way, you can kind of shortcut, that is run a perplexity prompt to like, build a kind of onboard and doc around how to prompt in V3 based upon all of the information that's shared from, like high authority figures. That's kind of what I kind of try to skew it towards people who have authority in that field. And then it creates like an onboarding doc and then I upload that doc and I say, whenever you're prompting for VO3, reference your onboarding doc. Do you think that's a good thing to do and do you kind of do any of that?
- [18:34]
C
Yeah, no, that's a really, really good point. And I've done a ton of stuff like that where I use perplexity to scrape, you know, the best resources for. Maybe I want an agent that's going to write HTML, or maybe I want one that's going to prompt V03 and other video models and stuff. But that's a really, really easy way to Leverage AI to leverage other people's expertise and just get that years and years of experience in your hands in a matter of seconds. So it's a really good strategy.
- [18:59]
B
Yeah, I like to think of it. It's like one of my favorite movies of all time was Matrix. And I always love the scene where he's like, I need to learn how to ride the helicopter. And it's like. And he just like learns to ride a helicopter. It's a little bit like that. Like, you should still learn these skills, but you can kind of shortcut to a pretty good outcome by just scraping data and creating onboarding docs for these agents.
- [19:21]
A
I think what you're also saying, Kieran, is focus your time on learning the skills that are really germane to your, like, business or your role. Right. Well, like, which is kind of what Nate's showing. You don't need to be a master prompt engineer for some of this stuff, but you do need to be able to have a solid workflow and understand your audience and your targeting and where you're posting and paying for your advertising so that when you run an agent like that, you'll get a good result. And speaking of good result, we just got a video back.
- [19:56]
C
We did just get a video back. So this one was like kind of smooth jazz music, which is why I smiled when I clicked on it. And you can see it's kind of slowly zooming in. And also what's interesting is we have words here again, right? And so we could basically tell it. We don't want any words on screen. We can tell it to have certain dialogue. You can also see that it ends off over here with like a black screen with a kind of a call to action. And this is where you could basically give it, here's my logo or here's the name of this product. And it could insert this kind of stuff in different areas based on what you want the advertisement to look like.
- [20:26]
B
One of the things I've been working on with VO3 is I am trying to create a like, three minute video clip. And one of the, you know, problematic things with VO3, which is interesting because store two has kind of solved this through their AI model. And we kind of talked about that on a previous episode, which is character consistency. Or in this case it would be product consistency. Right. And for what we mean for people who are watching this is like if you took a eight second clip that you have, but you wanted the video to be 24 seconds, or you wanted the video to be longer than 8 seconds, you have to create multiple VO3 videos stitched together. But you need the aesthetics of the product or the character to look consistent across videos. And sorto, you can do that for individuals because it's using the AI avatar of yourself that you create when you actually sign up to the app. Have you played around with like longer videos using VO3, like a longer ad that would need some form of consistency across product or characters or things like that?
- [21:25]
C
Yeah, so the way that I would do that is essentially you want to create your base source of truth files. So whether you have, you know, a video or a picture of your product that's just against like a plain white background. So it's just isolating just the image and then you can feed that image into all of these videos and then you could do the same with, you know, a character that you want to stay consistent. And so you're just giving it sort of like the source of truth. But no, you make A great point. VO3 is definitely a little behind on that aspect of being able to create these long running consistent videos in the way that Sora 2 now can do really well.
- [21:58]
B
Yeah, the way they solve the character consistency problem is really smart on Sora 2. It's one of my favorite things they did.
- [22:04]
C
Yeah, it's very, very cool. So I guess at this point, what I think I could show real quick is just what's actually going on in these three tools and why Nadin is so powerful. So we've kind of covered that this agent has a brain and it's autonomous and it can call on any of these six tools that it has. And for example, in this one it called on the Image to Video tool. And if I open up this tool, what you guys will see is it's actually a separate workflow that I built. And what's cool about this is this is an image to video tool. And now I can have this Image to video tool hook up to any agent. So I could build 10 different agents in NN and I could just say, okay, I already have an image to video tool. I just have to connect it here. And now if I ever want this agent to turn an image to video, I've already got that sort of like module built.
- [22:46]
B
That's cool.
- [22:47]
C
So it's very, very scalable, very modular. And now we have all these little reusable pieces that we use in our builds internally or builds for clients and stuff like that.
- [22:55]
B
That is super cool. Like you can build usable tools across your agents. I love that.
- [23:00]
C
Yeah, it's super cool because now I have, you know, like I made like a personal assistant agent that is one of my best performing YouTube videos. And I had like email actions, calendar actions, contact agent actions. And now whenever I build anything where I need an any email actions, I can just hook it up to my email agent. Because now the email agent's the one that has all the tools, it has all the prompting there's. And it just makes, you know, your, your time of building something way, way shorter. And it was already shorter in the first place with this no code software. But once you use it more, you're able to build things a lot quicker too. So it's very cool if you're watching.
- [23:33]
B
Along here, Nate, and you really want to get started, build your first agent, build some tools that you can integrate into that agent and do some of these tasks. One of the tricky things for the average user, like someone who's not really automation first or technical first, is where do I start, right? What do I even build an agent for? What are some ideas that I could get to automate things? When I was working in Zapier, that was like one of the harder things that users had is like they could think of maybe one thing where they couldn't think of five things. And I'm curious, if you were kind of given advice, how would you start on trying to figure out what are some good places to start to build your kind of first agent, your first kind of automation. And you're obviously a power user of anytend, but what tool would you use if you were just starting out? Like if you wanted to kind of take what you've done here and kind of get going and build some stuff?
- [24:29]
C
Yeah, it's a really good question. So when I first start to think about, you know, the first thing that came to my mind was don't run before you walk. And so I think when I see a lot of people getting into issues when they jump into a tool like eniden or make.com is they want to go straight to an agent because they see videos of agents, they hear the word agent and they think that's like, that's cool, that's what I want to build. And if you don't have a solid foundation or a good grasp on JSON or variables and the way that data moves from left to right in a workflow, then jumping into an agent is just going to be way too complicated. So really what I would start to think about is I want to build a simple automation first where I'm transferring data from one place to Another and I'm doing something, maybe I don't even need to use AI yet. And that's actually something that we've found with a lot of our clients was when we started to dive into some of their processes to understand like what would be the first project we worked on together. We ended up scoping it out and realizing this isn't even AI. This is just going to be a very simple automation. It's going to be a quick win, we can spin it up very fast. But more importantly, it's going to save a lot of time for you and your team. Because really when I'm building things, my mindset is how can I eliminate as much AI as possible. You know, AI is great, but it's a black box and it means that things are going to be a little bit more non deterministic. So if I was starting right now, I would think about what's something that I do every day, every week. And when I do that process I would start to write down the steps. So, you know, step one is this, step two is that, step three is this. And if I do those three steps in that order every single time, that is a great, great use case for something that can be automated.
- [26:05]
B
Yeah.
- [26:06]
C
So I would basically just process, map that out and then get into, I would say to go to end to end because there's so many resources out there, they're growing really quick and they're getting easier and easier to use. But if something like make.com looks a little less intimidating to you, then I'd jump into that. I want to say real quick, I think that some people spend a little bit too much time worrying about I'm going to take a gamble, which tool do I go all in on, which tool do I learn? But really I think when you learn any of these tools you, it's about the foundations, it's about figuring out what can be automated, where AI should go, when to use an agent, and if you can pass variables anywhere on a workflow, you'll be able to jump into any tool very quickly and apply your knowledge. Because they're all doing the same thing really behind the scenes. So I wouldn't worry too much about which tool you go with.
- [26:49]
B
Right, right. What is the most powerful agent you've built? It could be the one you've shown us. I'm just curious, like when you built, like what's the agent you built? And you're like, okay, I want to dedicate all of my time to this, what you've done.
- [27:00]
A
Right.
- [27:01]
B
You create Content, this is your business. What was the unlock for you that you were like, wow, this agent is insane?
- [27:08]
C
That's a really good question. So when you said the word powerful to me, what I'm thinking of is not what can the agent do and does it have the most tools? Because I've built something like this, we're not even going to dive into this because this is just, it's probably going to turn a lot of people off. But this is like a very, very powerful agent. It can take action in email, Google Drive, calendar, it has image, video and actions, it can post online, it can search, social media, all this kind of stuff. But when I hear the word powerful, what I think of is what system adds the most value or has, you know, had the biggest ROI for me or for a client. And really when I think of something that's powerful, I'm thinking of something that has saved me a lot of time or saved a team a lot of time. And those are very, very simple automations that are extremely linear and there's not any decisions that need to be made. There's very minimal AI and all it does is it goes, okay, step one, two, three, four, five, six, seven, eight. And every single time we run it goes step one, two, three, four, five, six, seven, eight. Because that cuts down all of the non deterministic decision making. And in automation, boring is beautiful. Predictability is your best friend. So really the most powerful system that we've probably built is somewhere in the sales process, a very simple flow where like I said earlier, a lead form is submitted, an agent immediately researches them, provides our internal salesperson with a brief or even just reaches out to that person, you know, shoots them a call or just shoots them an email and starts to nurture that lead right away. Because when you think about like where does money get left on the table? Where do leads slip through the cracks? It's a lot of times, because the speed to lead, you know, that's super important. And so if a business isn't getting back to their leads while they're the hottest, while they're most ready to buy, then you're losing a lot of money. So I talked earlier about kind of the flywheel of a scalable system. If we can build a system that in turn grows the business, as the business grows, the system is going to get used more and the system starts to get used more and more without the team even knowing it's getting used more because they don't touch it at all. A system like this or personal assistant, which is what people want to build right away is they always say, I want my own personal assistant. That's great because you know, hey, move this meeting there or get my emails from this or whatever. And that might help grow the business because maybe you can be a little bit more productive on a day to day. But as the business grows, are you going to be using that personal assistant more? I don't necessarily think so. You know, it doesn't scale the same way as some of these linear automations.
- [29:33]
B
Right. I think that's true. I think that's really good advice as well for where people in terms of where they could start. I think you said a couple of really important things which is, hey, like what do you do day to day? Like write down what you do day to day and you can just give that actually because I was showing someone how they could do this, you can actually just give that to ChatGPT and it will literally build you a table of like workflows that you can build. And actually you can ask it, you can say, hey, like I'm going to use this tool. It will build you tutorials. If you said like I'm going to use N8N, it will give you tutorials in N8N to build each of those workflows. Now we suggest you go to Nate's YouTube channel because you'll have better details than ChatGPT but like to get started. And then to your point, I think what I'm seeing is like there is a whole cult of N8N workflow. The bigger the better on X and these social channels. And so to your point, I think it's really good advice to start with something small, get a quick win and go from there. And so I think this is super cool, like very, very cool.
- [30:31]
A
And Nate, for everybody watching, are you going to share this sample, like basic image workflow for everybody? I think you are right, we'll link up to that, That'll be down in the comments. And so if you're just getting started with N8N and you want like a early building block, what Nate's walked through here, you'll be able to go and basically kind of run some experiments and see how does this stuff work so that you could then maybe internalize and think back to what Kieran was just saying. What are the problems that I have where that good consistent automation could really solve them? And maybe you need AI, maybe you don't. I think that was a really valuable point that you made Nate, that sometimes automation without AI is still very, very valuable.
- [31:10]
C
Absolutely. It's where I think a lot of businesses are now starting to get into the mindset of automating things because the word AI was put in front of it. And then they realize how much opportunity there is to just do some basic data transfer that really, really saves a lot of time. And I think one more thing I wanted to say that would be probably applicable to a lot of businesses and businesses right now is a good question to ask is, you know, what sops do I have? What Standard Operating Procedures do I have? Where on that document it probably explicitly lists the 10 steps in that order and it shows, you know, who's involved in each step, it shows the data source from each step, and it shows the transformation that should happen in that process. And that's a really, really good place to start because typically, like I said, that's a very predictable process because if you've went far enough to build that document sop, then that means it's probably happening frequently enough where people need to refer to that. So that's a really good place to start as well.
- [32:01]
B
Very cool.
- [32:02]
C
Yeah. So I wanted to touch on one more thing about the outputs that we were receiving from this agent. This is a very, very general, high level. Like I said, it was made to be a template for everyone in the audience to be able to just play with and see how it works. And this agent doesn't know that it's always making images for, you know, VFX ads or UGC content, or maybe it's making, you know, stories. It doesn't know what it's doing. And so earlier when I kind of alluded to the importance of prompting these AI image or video generation models, when VO3 dropped, we were seeing some really cool stuff and people were like, how did you make that? And it's all in the prompt. So I found a few examples on X that I wanted to show you guys that you may have seen when this whole thing dropped. But this is an example of a really, really intense prompt here. So you can see there's a description, there's a style, there's a camera, there's very specific elements that go into the way that the AI interprets the request. And now what we see here is this eight second clip where I think that Ikea would be more than happy to put this in a commercial.
- [32:58]
B
Yeah, this is incredible. So good.
- [33:00]
C
Like, that is insane. And I found another example, like one for Apple, like an Apple watch, you know, like, this is the kind of stuff that would take a long, long time and probably a lot of money to actually create. That one was Apple. This one looks like it's PlayStation. Like we have all these different ways that we can use prompting to make these AI models really, really specialized to create things, you know, like that, that Jordan shoe, that Nike shoe. There's just so many different things you can do here. And it's all about the prompting. So the reason I wanted to throw that out there is because maybe you looked at the outputs we got here and you were like, that looked awful. First of all, you know, the big upgrade was the fact that we have like these consistent things where now I can take a picture of cologne in my hand and it can turn into an actual video. But you get in here and you make the prompts better and better. And also like these videos have the exact prompts. You can kind of copy that structure exactly. And now you can, you know, just create some really, really cool stuff.
- [33:48]
B
Yeah, 100%.
- [33:50]
A
And we just did a show with Rory Flynn and he walked us through a bunch of creative workflows using Weavy AI, which is a really cool creative workflow tool which kind of builds on what you were just showing, how if you really can build specific creative prompting, you can get really remarkable high scale creative results.
- [34:07]
C
Absolutely, absolutely. I remember when I first started doing end and stuff and like my very first freelance gig I got was building an automation to help this person automate their LinkedIn posting. And so, you know, every morning was going off, it knew about their business, it was researching the industry and it was making these LinkedIn posts for their approval. And I remember them asking me, could we get some images put in here, like AI generated images? And I was like, no, it is not there yet. The words would be all jumbled, it would just look awful. But now we really are at a place where I believe we can do that totally. And you can see how simple this is. You know, someone would submit a form of what they wanted, the agent would do research and create the post. This agent was trained on specifically building prompts for LinkedIn graphics. You know, like maybe there's a little icon and then there's like some text. And then actually this was using OpenAI's image generation model, but it was that simple. And these agents, you can see they have no decisions to be made, so we're really, really keeping them on the guardrails that we want. And this is exactly what I mean by step one, step two, step three, step four, step five, step six. Every single time it happens in that order. And this was a really, really powerful system because now that business doesn't have to do any more research. They wake up and they have three posts with graphics that go along with those posts and they can just approve which ones they want to actually go out.
- [35:19]
B
Love that. That's such a great simplistic example that everyone watching is going to get the value of immediately.
- [35:25]
C
Totally. Yeah. And we could also link the video. This is a free template as well that people could play around with if they want.
- [35:30]
A
Please. We'll link both of those up. Nate, you've given us the ability to create awesome image or video ads, showed us the basics and fundamentals of building simple and less simple agents using N8N. And I think it should be an eye opener to everyone out there that way more is possible with automation and AI when they work together than probably most people understand to solve their day to day problems. And we recommend just going and playing with some of these and applying them to your own problem so you can actually see the value that you could be getting by kind of adopting and learning the skill.
- [36:06]
C
And I think that there's a lot of value in just, you know, exploring with before you even think about automation. Like a lot of people that I talk to don't even know what perplexity is.
- [36:14]
B
Yeah.
- [36:14]
C
Or they don't understand what Gamma is. And like all of these tools that take no, absolutely no technical knowledge to get into. Like you could actually just go to like Google Gemini and start playing around with images and videos there and then the unlock is when you understand. Now I see what it can do. But if I can have this running automatically all the time without me talking to it, it just unlocks a whole new world. So it's a really good point there. Kieran, anything else you ready to wrap?
- [36:40]
B
No, I'm ready to go play with some AI workflows. This is super cool. Gonna go play with all these templates.
- [36:45]
A
This has been awesome. Thank you, Nate. Thank you for everyone watching today's show. We'll see you next time on Marketing against the Grain.
- [36:50]
C
Awesome. Thanks.
- [36:54]
A
Sam.