Marketing always needs more: more content, more personalisation, more variations, more testing, and all of it faster. AI is the iPhone moment for that pressure, turning zero into the new sixty percent so teams can start much further along and let people finish the last mile.
Thibault Imbert builds an entire campaign live, inventing a Scottish beer brand from ideation and naming through product photography, stock imagery, ads, short videos and a working website, all with AI. His practical thread is that humans are bad at prompting, so you should show rather than tell, using mood boards and reference images to get exactly the vibe you want. The result is a hands-on look at "vibe marketing" and a clear-eyed take on where the tools shine, where they still slip, and why he sees this as a bike for creativity rather than a replacement for it.
Auto-generated transcript - may contain errors.
Tap a timestamp to jump the video.
Hello, wonderful people. Thanks for having me. My name's Thibault. I run product and growth at Creatopy. A little bit about me. Spent years in my career oscillating between design and development and teaching and being a sales and product manager and a growth marketer, and so this, I think, is interesting kind of like set of experiences with what's happening right now with AI and marketing, and it's been helping me understand how we can leverage AI for marketing.
I work for very small startups like Microsoft and Adobe, and some very big public companies like Pixar and Creatopy, and it's been a fun ride. And so today, I I actually had just started at Creatopy, which is a company focused on it's a platform for managing your campaigns, creating your your campaigns, and managing the automation of your campaigns for the creation, the publishing, the insights completely automated with AI.
And one thing that we wanted to do for you guys, if you want to check it out, we're giving you this discount, three off from the Plus subscription, so free swag for you guys today. So you'll have the slides so you can remember that later if you want to get access to it.
So we're gonna talk today about quickly about adoption of GenAI quickly, and I don't need to convince you that, obviously, it's changing everything in the way we work. But I wanted this talk to be extremely applicable and reusable so that you can take this tomorrow with your team and start using it.
So it's gonna be, we're gonna go deep with the workflows and see how we build things. So hopefully you're gonna be able to take that away and reuse that with your team. I think I was at the speaker dinner yesterday, and we were talking and acknowledging that, for those that are old enough to remember, before the iPhone, this is the iPhone moment for AI.
I think there was different waves of disruption. First one with them was before the iPhone, after the iPhone, and there was before AI and after AI. As a reminder, I'm sure you need to convince your boss or anyone around you that AI is here to stay and it's being adopted faster than Internet or PC.
ChatGPT, fastest adopted and growing product ever in the history of software. And we're here to talk about marketing. And marketing, there's always more content that needs to be created faster. Demand is forecasted to continue to grow, and you're gonna be tasked to always create more, more personal, more personalized AB testing, better performance.
And guess what? AI can help you for that. So just for context, Figma just announced just shared a AI report a few days ago, and it was interesting to see that half, in terms of adoption, AI is more broadly adopted amongst developers than designers, almost half.
You can see sixty percent amongst developers use AI, thirty percent for designers. And the reason is simple is that models are really good at text and they started there. So generating code, when I worked at GitHub, we were actually the first enterprise scale software using AI to generate code because it's text.
But doing it for design, for text, for layouts is much harder. But it's moving really, really fast. I believe that in the next six to twelve months, we're gonna see the same level of adoption, and designers are gonna be using AI as much as developers do.
So there is still some skepticism, and I want to show you and be transparent with you on the limits of how much you can use AI in your marketing, but show you also what you can do tomorrow and start using it to make your team more productive and more efficient generate more revenue and more business.
So it's a vibe. Who has heard of vibe marketing? Yeah. So vibe marketing, there's vibe coding that was coined a few months ago by one of the OpenAI co founder, Andre Karpathy, who was talking about vibe coding. This concept of being with the large language model generating code, and you're kind of like going with the flow instead of like going to Stack Overflow and searching for things and like doing things very manually.
You're kind of like vibing and letting the model take you to places, and then you're And this is vibe coding. It's a thing. And we're seeing vibe marketing being a thing. And this new approach helps us marketers ideate, jam much more rapidly, and I'll show you.
I mean, I've been in situations where as we're jamming, as we're talking with the team, people, as we're discussing, are already designing and generating things. You'll see an example in the slides where actually, Ollie, hey, one of my friends and former team member at Pixar, we were talking on WhatsApp, and I was like, look, I'm doing this, and he was already building the website as I was speaking using some models.
Another thing before we get into the tools, I think it's a bike for creativity. If there's one thing you want to take away is that AI is now making basically the zero the new fifty, sixty percent. We used to start at zero. Now you're going to be able to start and get you very quickly to fifty, sixty percent, or seventy percent, and you're going to be able to use really your team to finish, get from the seventy percent to one hundred percent.
I do think in the next month or even before the end of the year, we'll get to eighty, ninety percent, but this is extremely powerful. It's bike for creativity and productivity. It's not gonna be replacing every job, though it is going to replace some jobs.
We've seen that. But for design and marketing, it's also gonna be making people, give superpower to people. Alright. So let's get into the specifics. I'm gonna walk you through a campaign we're gonna create from scratch, and we're gonna be using AI, and everything you're gonna see here is gonna be using AI so that you can do some vibe marketing.
Are you guys ready? Yeah? Okay, cool. So I was thinking I need to let's pretend I wanna build a really cool Scottish beer brand. Probably I should have picked whiskey, but I was like, okay, love beers, so I was like, can we create a brand, and let's pretend that a customer wants to, you know, we wanna kind of pitch that concept, and we're gonna go from ideation, brainstorming, stock photography, product photography, generating ads, and creating the website all with AI.
You can, if you know the tools, you can do that in an hour, and you'll have the day you can reuse the prompts and you can use that for your business and reuse. Okay, so first brainstorming. Just for context, at PixArt, we did a brand campaign recently, and this is the final result of the campaign.
These are videos and these are real people. But when we were brainstorming, all of the mood boards were mid journey generated, and through the streaming and Zoom and stuff, was like, wait, have we shot the thing already? No, no, no, this is all mid journey, and so all of that is AI generated.
You can see it's like a year old, so you can see it looks a little bit AI, but that's already, like, I'll show you some examples where you can tell the difference. But we were already using AI for all of the ideation, brainstorming, and trying to already picture the story.
And that's actually really cool because you can make a point and you can go much faster in making a point by showing rather than telling. Like we say, a picture is worth a thousand words. So let's, for instance, talk about what Midjourney has done recently with mood boards.
Across the entire presentation, the core approach is to show rather than telling the models. Humans suck at prompting. Right? So I personally do not believe prompting is the end of the road. Yes, it's fine for localization or conversing for different topics, but when it comes to design, most humans can't describe something like a designer would.
So what do you do? You show. So for instance, my dream is to retire and open a citrus farm, and I wanted to picture that and try to convince my wife to really retire as soon as possible. And so if you do that today in any large language model that supports image generation, you're gonna get something average, one size fit all, generic.
So Nelligans Citrus Farm, my hope is that there's a bistro in there, all that good stuff. But I already have a Pinterest board of the vibe that I want. This is I want metal, I want wood, and I want blend with nature. How do you describe this in a prompt?
Impossible, but you can show. So now in Midjourney, you have support from mood boards. So you can create a mood board, you turn personalization on, you upload your images. Sadly, you can't connect to Pinterest, but it would be really cool. I uploaded my images, same prompt, result, different result, but exactly the vibe that I want.
All I had to do was to reference in my prompt the mood board, and it's all built in mid journey. You can have you can play with that tomorrow. All of them showing here is not something, you know, exclusive or like private access or anything.
All of that is accessible. Okay, so if I can show rather than tell, can I use that for all of my campaign ideation, brainstorming, vibe marketing? Absolutely. So let's vibe. Let's find a cool name for the beer, CHA GPT four point zero, I guess the default.
And it's coming up with some cool names. Kiltswinger, Whibisty Ale, Piper's Pint. I really like lock hops, and I'm like, oh, lock hops. And so I'm like, give me more. And we land on the hopness monster. I have to admit, I was like, that's pretty good.
Seriously, like, I was like, that's really good. My friend Ollie was like, dude, it's really good. You have to do this if, you know, everything fails. Go launch this beer. So I actually asked, okay, can you generate like the label of the bottle?
And one thing that's really important to know is that GPT and the image models historically were terrible at image generation. You've probably tried this. Who has tried to generate images in chat GPT? Right? You're like, no, thank you. And this is what's Slop AI.
So here, how come this is so beautiful? Like the text rendering used to be a nightmare. It's actually looking really, really nice and polished, and it's generating more. And I still think Hotness Monster is really good, but this is what AI image generation used to be, like a year ago.
Well, there are lots of models like Ideogram and Recraft and OpenAI that have released new models that are extremely good at marketing. Ideogram, we've used it at PixArt for text generation and logo generation. Recraft is also incredible. But OpenAI just released recently GPT image model one that has kind of like, you know, reset the baseline and upped the bar for everyone.
And so my, I would say, as we speak, this is the best image model. When I say best, mean there are many, but everything I've done here is using GPT image model one. This is really, really good. So as any good beer, I'm gonna jam, I'm gonna do some more vibing and say, okay, what are different beers that I want?
Obviously, I wanna have an IPA, a Hefeweizen, a pale ale, and different colors. And so GPT is now generating the different variations for me. And I'm more a Hefeweizen. Love a good Hefei, like the colors. This is pretty cool. And so now I wanna find some inspiration on Pinterest about my staging and what the product looks like.
Again, how do you do that? Well, you can show the model. So I find this image, and I like the color. I like the reflection, and I want the model to do exactly this. So what I'm gonna do is I'm going to add that as a reference.
There's no concept of mood boarding in GPT four point but you can just attach an image and say, I want my image to look like this, and it will generate exactly that. Now, at this point, I mean, for those I mean, those of you you probably know all this.
This would take a long time, even for an experienced designer to put this together, like go to Photoshop and do all that, or maybe in three d with three d tools I'm running, but it wouldn't take ten seconds. It would take more time.
But you are experimenting, ideating, so you don't want to spend so much time, and now this is giving you those results extremely quickly. So now my beer empire is starting to look really good. When I was rehearsing the slides yesterday, the person was like, wait, is that is that a real?
I have never seen this beer. No, it's not real. But it made me really happy that that the they thought that it would be. So now we're gonna add some foam. We're gonna make it a little bit more, even more real, and one of the things that you wanna do is use GPT to give you prompts, and this is something I discovered recently.
Again, it's really hard to do prompting, and the models themselves are trained on their own prompts, so you want the model to give you the prompt. So instead of saying, add some foam and dripping, like, say, give me a prompt for image model one to do this.
It will generate a prompt, and it will now generate the image for you. And you can do that in batch. Now, notice that one of the thing that happened is that my label has been modified. And this is something that happens a lot with It's not technically hallucinations, but sometimes the model do make mistakes.
It's okay for some jamming with the team, if you want to present that to a customer as a kind of like direction, like a semi high fidelity prototype, you want to fix that. My Nessie is much cuter with little hop hat. I don't know if you noticed, little hop hat.
Isn't that really cool? It's gone, and so I want to fix that. How do you fix this? You can actually provide the image of the label that I've saved, and you can say in the prompt, use the close-up provided as the third image to accurately incorporate the label details.
So you will basically realign the model and force the model to integrate that, and that allows you to fix it. And voila. Okay. So now, are we vibing? Yes? Yeah? You're liking this? For people that don't drink alcohol, I'm not saying you should.
And for the people that don't like beers, next time I'll do whiskey or something else, or just mineral water. Product photography. So now you're thinking, this is great, Thibault. How do I make this look really, really real? And same thing, I'm I wanna stage.
I really like this image I found on Pinterest. I like the the little how do you call that? The little bubbles of water? How do you like the Yes. The bottle. And I like the light. Impossible to prompt this. So what are we going to do?
We're going to put in the prompt. We're going to say, hey, I want this to look like this. And voila, now you can see the light. I can see the reflection. It's I'm really thirsty now. And so we're gonna continue, do that for all of them.
And again, this is infinite. If you like, if you wanna even put it, Ollie had my beer drunk by James Bond, you can do that. Be careful with copywriting, obviously. But you can replace stock photography if you want to put that in the hands of people.
You can in seconds. Same thing, you provide an image as a reference and it will create variations. Now this is extremely powerful for marketing because for AB testing, tools can let you do that. You can do that manually. In Creatopy, we are actually gonna do that with AI soon, but very powerful for testing.
Different backgrounds, different vibes, different types of people. You can do different, like, culturalization. So let's talk about stock photography since we're talking about putting that in context with humans and stock photography. Well, which one of these photos is AI generated? Well, all of them.
And all I had to do was to basically give an image of, like, the type of scene that I want, and I gave me a prompt, and I said, give me different colors of the jacket, different time of the day. And I got these results.
And, what I was able to do here is if I wanna integrate the so I don't know how to pronounce it. Ur hold on. Holkrot? Urquhart. Urquhart Castle, which I've been here, by the way, last year, three weeks north of your in the Highlands beautiful place.
And if I wanted to integrate that in the background, it would work. Now, technically, it's a little bit provocative. Some of you might notice, oh, this is not exactly what it looks like, but you get the idea that you can, again, for the vibe and for the style, you can actually get the team to see, oh, yeah, that's what it would look like and get your point across with the vision of the campaign.
Now, you can generate the same image but, again, create infinite variations, and if you want to add a little bit more, historically, models were bad. Your skin used to look like kind of like Barbie, little bit like too much like clean skin. Skin needs to have a little bit of noise, imperfection to look real.
Now the models are doing that really great, but you can also use tools like Magnific dot ai to add a little bit more of that noise to make your pictures look even more real. You can see the before and after with the compression.
You're probably not seeing, but you'll see the before and after where the face looks a little bit even more real by doing post processing on your images with a tool like Magnific dot ai. Okay, let's now turn that into ads for advertising. And I'm gonna come up with some cool copy for ads.
And I have to admit, this is again a moment where I was like, oh, this is really good. Yesterday at dinner, we were like, no, it will never replace humans' creativity. I'm like, there are some really good moments where I'm like, this is better than what I would have come up with.
But you can see short taglines where myth meets malt, Scotson's newest legend in a bottle, a lox sized flavor. All of them are good, and so I need to now I have these images, and I need to make space for my text. You can use generative expand tools in Photoshop or other tools to make space and make, basically, extend the image to the left.
You can invent pixels, which is really a pain in the butt if you don't have such tools because you would have to regenerate and specify the aspect ratio. Well, now you can just use these tools, add your different taglines, and now we're getting into more of like a vibe of the different performance marketing ads, what they would look like, and see the direction.
So I wanted to visualize our country launch, so I did that on obviously bus and in the city. And yeah, this is allowing us to picture what the campaign would look like, and you could actually do that in any type of environment, whether it's in Scotland or other places.
If I wanted to create variations, I can also do that. This is actually done with Creatopy where you can bring your assets. We remove the background for you. We create, we already have the templates for all the different size, billboards, skyscraper, all the different ratios, and you can just put the text.
We automatically change that everywhere. You click a translation button. We do the translation, and it's ready to be published on your ad channels. So this is kind of what it looks like in the product. Now, how do we make things move? So if you're thinking, wow, this is cool for static images, you can also do videos.
So how many of you know that you can turn images into videos? Okay, well, you're about to discover this. So there are models like Runway or Google VEO two, Cling dot ai, that allow you to take an image and turn that into a video.
Now, of course, this doesn't work for anything longer than ten, twenty seconds where there's gonna be some artifacts, but for five, ten seconds, it's really, really good, and in my test, VEO two has been one of the best, but, you know, you have to experiment yourself.
But in this example that I'm gonna show you, I've used v e o two. So how do you prompt an animation? We are gonna ask the model to give us the prompt. So just ask the LLM to generate it for you. So in this example, I am going to say, give me four prompts, and this is something I use all the time.
Give me four prompts to do this, and GPT or any LLM you wanna use, whether it's LAMA, Mistral, or anything you like to use, will give you different. Now here, it's good to use the same model that generates images, so in the context of GPT four, because it's its own image model, it's better to use the same LLM model for the than basically the same image model that belongs to the LLM model.
And here it's giving me some results. I pick one, which is to animate a few birds flying softly in the sky over the Loch Ness, and that's it. That's pretty good, right? I mean, again, think about it. Ten seconds, fifteen seconds. I have that.
Now, I have So this is laggy here. I don't know with the PowerPoint conversion. It's not smooth, but in original version, it's really smooth. And you can see I have a version where it's dripping, there's no birds. There's a version where the light is changing.
We're vibing. We're vibing. So this also works for real products. If you're thinking, okay, Thibault, this is great. This works for a fake product. How does that work for a real product? This is here a candle shop. I just Googled, found a store, London based.
It's Octo and Co candle company, beautiful candles, and this is one of the image on their store, and I was like, wow, let's bring some life to this. So I said, I put that image, I said, can you generate me some prompts for slightly animate that so that could turn that into a story?
And it did. And it actually, you know, added the bergamot citrus flowers with the wind. It I don't know if you noticed the flame flickering. It's just incredible. And again, it took a few minutes to do that. Maybe not maybe someone might say, oh, you know, the flowers, maybe not super realistic.
The end yeah, but again, we're sixty, seventy percent there. Okay. Last but not least, now we want to talk about the website, and thanks, Ollie, for doing that because I was building the deck. I was like, oh, this is so cool. Like, I'm doing this.
Check it out. And I was like, oh, I think, you know, you should build a website. He's like, yeah, it's done. Here's the website. I was like, thanks. So we need a website. You can generate apps with text. So we've all used GPT, text to text, multimodal text to images.
There's also models to generate videos. I didn't go as far as generating audio, but we could generate audio music entirely with text. And there's also now text to apps, text to website. And so one of the product that's really hot right now, it's called Lovable, based out of Stockholm, and it allows you to basically provide some context, put a prompt, and generate an app.
And it's not just like like, you know, like static. It's like responsive, like like fully what a developer would generate, and a lot of companies are taking this, and actually it's sixty, seventy percent and taking that to production. There's already companies that have website completely built with Lovable.
And so that's what we use here, and that's it. In a few seconds, generated the website. Have you spotted the hotness monster? We added in the prompt to make it a little bit funny. It integrated the pictures. It's got everything. It's got all the, you know, the fields, and if I wanted, I could now take this and hook it up to a Mailchimp, and that's it.
I can start, you know, getting subscribers for when beer is when the hotness monster is unleashed in Scotland. I think that's Personally, I think that that's crazy, that we can do all that so quickly, and this is all the tools that I used for this.
This is the moment where people are like, yep, that's the only slide I care about. Thank you very much. But all the slides, I'm gonna post them on LinkedIn, and you can get access to them. So what's the bottom line? The bottom line is that all that is possible, and there's this article, interesting article or interview on on YouTube from this company, Systematic agency, marketing agency based out of the in the US, and they're talking about how such a workflow that they've adopted already has allowed them to reduce the time
they're spending per project, not to fire people because they can do more with less. They can basically take on more business. And I remember talking to a head of design who told me, I wanna spend time, more time on the things that I love with the team doing, the ideation, the design, and that that kind of vibe.
I don't wanna spend time with the doing and the automation and all that stuff. No one likes to do this. No one likes to generate fifteen variants of something. Give me the different sizes. You can use robots for that. But they were able to basically do more business and maintain the same quality, So it's not Slop AI where, yeah, you can do more business, and your customer's like, yeah, this looks like crap.
So what does this mean for us and for marketers if you have a team or if you work in a company, want to adopt that? I think the first thing is that the barrier to adoption creation and production is much lower. It's really, really powerful.
I've had people in my team, product managers, saying, I'm not really good at design. I need to take a class and course on Figma so I can make my point across. No, you can now just do it. You can there's tools that let you generate literally a Figma.
You can also spend more time thinking, jamming, really keep humans focused on the true creativity that we need. So for me, the good analogy I've heard is it's a bike. It's a bike. It's a bike for creativity, and I think that's a beautiful analogy.
It will, yes, it has replaced some jobs, but it's gonna give superpowers to many of us to hire more people, to be more productive, and do more beautiful things. It's a vibe. It's vibe marketing. Thank you so much. The slides, thank you so much.
If you're curious and you wanna get access to the full slides, I used to upload that somewhere, I don't even know now, where I don't have a Dropbox anymore. But I'm gonna send it on LinkedIn tomorrow. This is my profile if you wanna follow me.
Don't worry, I won't ask you to comment to get access to it. This is so annoying. What's the I know why people do that. It's so annoying. And so, no, I won't ask you to comment. You'll get access to slides directly. But thank you so much for having me.
I have another, I think, roundtable later. And you so much. Bye bye.
