[Music] [Yuji Nakagawa] Morning, everyone. [Veronica Campbell] Okay. What's up, Las Vegas? I know, it's very early in the morning. Bear with me here. And you had a late night at Bash and I'm jealous of anybody who did the simulator and the pit because I was too late. Yeah. These are the folks that are committed, right? I know, right? Very committed. But we have such an exciting project to show you guys, so thank you for being here. Thank you. Yes. So we are going to take you a little bit through Doritos journey. But first, we're going to start off talking about who we are and what we do. I'm going to go through what the PepsiCo Design Center is and how we've adapted AI. And then we're going to use some of our projects that we're leveraging specifically Doritos, but also a fun Gatorade project. And then how we've adapted custom models. And then Yuji's going to take it away and kind of give you the best practices, how you can implement this in your role. And then how we're future-proofing Firefly custom models.
And my name's Veronica, but I'll talk about myself later. I'll hand it over to you, Yuji. My name's Yuji Nakagawa. I work for Adobe Professional Services. I am a Content and AI Solution Architect. - Go next? - Yeah. You want to take it? Here you go.
So if you think about Gen AI, it continues to be a major topic.
A major topic of discussion and hope for driving business outcomes, right? From cost efficiencies to outcome outlets.
The hope for GenAI is real. Marketing in particular, I think, there's the biggest opportunity for injecting GenAI, right? And with all of the announcements, the news, the discussions around GenAI...
Yeah, it drives a huge amount of hype, right? You guys have all seen this. At Adobe Professional Services, we integrate revolutionary technology into your everyday workflow. We turn hope into reality. We drive the hype into the how. We do this by leaning in. We stay ahead of the curve. We work with our products and we know about the capabilities well ahead of the GA, right? And we also drive best practice insights, right? Our Adobe Professional Services differentiator is that we focus on Adobe technology.
Our differentiator is that we have hands-on keyboard experience across all industries, across all geos. We have consultants worldwide. And we do this by driving collaboration and innovation, right? We partner with our product team and our customers and we tackle readiness issues and we help drive collaboration and formulation of best practice. And so speaking of best practice, we're going to get to that. But I'm going to talk about innovation or I'm going to hand it back to Veronica. Yeah, thank you so much. And you guys have been so helpful with us. - It's been a great partnership. - It's been awesome.
All right, so I wanted to start off a little personal and at the Design Center, we have this thing called a who's who's board. So I wanted to share it with you. It's kind of to get to know your colleagues a little better personally. So I am born and raised in the land of Bruce Springsteen in Bon Jovi, and when I'm not here at Adobe Summit, I like to go to the car track. That's why F1 was so special to me. And then also traveling and also karaoking with my colleagues. And actually, on the bus yesterday, I think a lot of people like to do that as well here.
I've always had a niche for drawing. When my mom, she raised me as a single mom and I'd have to go to her shifts and she just threw printer paper or an eraser board at me. And I just really enjoyed illustrating and just like creating in general. And with that, I kind of wanted to just learn to create more. So I'm going to date myself and I started drawing on Microsoft Paint.
I feel like this is the right crowd.
Coding on Myspace, making my page look amazing. And then even like After Effects, I wanted to learn, some 3D, and then now AI is kind of my new project. And it's all about being curious and I think that's what AI, when it's used successfully you have to have curiosity.
My career started a little different. I was working on album covers at Island Def Jam Records and Republic Records. Done covers for Shawn Mendes, Noah Kahan, Bon Jovi. But that's a very niche place to be. You're making art for artists. But I wanted to learn more about what consumer buys as a brand, and what people really love and dedicated to the brand. So I moved to PepsiCo about-- Oh gosh, 2018. And I started actually illustrating. So I went back to my roots for Mountain Dew doing limited time offers and customer specials. And then PepsiCo Design has so much to offer. So now I'm on the Global Foods group overseeing Doritos and Cheetos, making sure everybody has a consistent brand look and feel for our guidelines.
Mauro calls us unicorns. I want to tell you more about the unicorns here at the Design Center. For the past 10 years, Mauro Porcini and our CEO, Ramon Laguarta, have built an amazing capability. And we're here to create joy for consumers and leverage design as a secret weapon for business and innovation and setting PepsiCo apart and making it a really a design driven company. And I love this photo because we just have such a unique group of people and they all bring their true selves to work. And it's just-- It brings such a great culture to our team. And just to show like all the amazing stats, it really comes out on here.
And we kind of just want to show all the multidisciplinary approaches we have to the Design Center. So we have brand building, so building a cohesive brand experience. We have innovation, so they work a lot on product innovation and design thinking. And then we have corporates operations because designers are all over the place, and we need somebody to keep us intact.
But these three segments give such a diversity to the Design Center. And now with this disciplinary approach, we really have to emphasize that when using AI.
So I don't know if any of you went to Adobe MAX as well. I know this is a very different conference than that. But we did have this talk about how we are adapting AI into our workflows. And I thought it was just really good to level set with a good old corporate America graph.
I'm not going to get you too in depth into this, but I wanted to show you kind of our AI adoption and our people first journey because it really has been a journey for the past three years. Yeah, please take a photo of it. We also have that on the Adobe MAX website. I think everybody can see it. So but just to like give you a high level, we have this circle. It's ever evolving. There's no start to finish, but I'm going to start with bold implementation. And it's all about partnerships. You want leaders to support you to unlock more capabilities. You also need a good partner like Adobe. They have been amazing to work with. And then also you need to be best friends with your IT partner because they give you all the new technology. Shout out to Dre. I don't know-- Oh, okay.
He is the one who helped me get here to work on all this cool stuff.
And then we have training is all about education.
Gathering insights and getting a base level of where your team is at and then creating a syllabus off of that.
And then empowering creatives. It's all about communication. I say have an open mind when people are coming to you. Surveys are great. Personally, I like the real talk of one-on-one and getting the real feedback about the technology if they're finding it useful.
And then last but not least is strategy and practice. So we want to find a project that supports these big wins and unlocks more opportunities with AI.
So with that, I'll go into some of the practices we have made successful and one of them is Gatorade because I wanted to start with that one, which is great. It is the first GenAI platform on a physical project using a natural language model. And I think it launched, what, five months ago, and there's almost 100,000 generated by consumers. So truly amazing. You all using Firefly and here's just like a cute little sizzle reel to show you a little bit more.
Yeah. All right. Now onto the more fun stuff. That was amazing. Thank you. You did that at MAX? Yes. Yes, we had it at MAX. We actually had a food truck at MAX or kind of food truck, but with all the bottles, just a cool looking truck. And the lines were out the door of Creative Park. They had to shut it down at some points. It was truly a great experience and-- It was the most popular. Yeah, Yeah. Yeah. So hopefully, next Summit it will be here. And I hope there will also be a Doritos food truck here as well. That is my manifestation for next year, but yeah, let's get into Doritos. So I hope all of you know what Doritos is. Does anybody have a favorite flavor? [Woman] Cool Ranch. Oh, why don't you give her a Cool Ranch one. Come on, throw it over here. You get a surprise, yeah, if you participate this early in the morning.
You get a Doritos bag, you get a Doritos bag.
Thank you.
There we go. More fun stuff to come. So Yuji has some good interactive stuff. But just in general, from a marketing perspective, from a brand perspective, we have spent plenty of time figuring out what is the look and feel of Doritos. And we've gone from just-- We call it our brand pulse and all this graphic and even I think the sound base, even Sonic branding.
And I think that everybody sees a triangle and we just want to like, capitalize on that. When you see a triangle, we want you to think of Doritos because it's a symbol of crunchy goodness, a beacon of bold flavor and a statement to celebrate diversity. And also, we're kind of a brand that's not straight on. We like to live a little bit on an angle, so this was kind of the solution for that. And now as fun news and trying to push our more bold flavor, we're doing a culinary platform called Doritos Loaded globally. Oh, that kind of cut out a little bit, but you get the gist of it. You see it all.
But we're flipping the chip on all expectations going beyond the bag to create all this new stuff with a delicious base. So whether it's pizza, chicken fajitas, and we had century egg come in, which is a new one. Whatever you want to do, whatever you can pull out of your fridge, you can add Doritos to that. Anything to ignite bold self-expression.
And obviously, we're using the triangle at all times, but this one we have to be a little bit more flexible. We have so many touch points as you see with food trucks, out of home, point of sale. So making sure that this is really super hyper-personalized.
And not only that, this is-- I know like there's some TikTok trends going on when people cut the bag and use it, but for other people, this is a whole new behavior that we have to educate our consumers on. So we wanted to make sure that we have the proper photography and just making sure that it looks delicious and people understand what Doritos Loaded is. I remember people being like, "Oh, it's like a pizza flavored Dorito chip." I'm like, "No, no, no, no, you put the pizza on it." And we actually had a few trials. So we had UCL of last year, tasted Dubai, Winter Wonder Park in the UK. And we got a lot of learnings from that. And with that, we had a whole pre-production of good art direction. But overall, we want to embrace imperfections, make sure it looks real fresh ingredients and, of course, the lovely triangle. Are you guys sick of seeing triangles yet? - No. - Oh, okay. Well, I have squares on this one.
So yeah, well, and we have so many-- We only had a limited amount of recipes that we can create. And we know that there is a need for it that I'll go into, but luckily, PepsiCo and the Adobe partnership, we have Firefly custom models to help us with that. You Know, we should pause and we should talk about this. - Yeah, go for it. - Right? Because ultimately, we're here to learn. - Yeah. - So take us through this. We're looking at, on the right hand side, you see like the subject. I think those are all what you trained on, right? Mm-hm. And we're going to talk about best practices in a bit, but you can see there's some consistency going on, right? And you want to make sure you have that. And then in the middle, those are the generated outputs. - The output, yeah. - Right? And ultimately, when you have a very good training, it's going to lead to good things. It's going to lead to very good outputs, right? And then on the left hand side that's what the best practice we're going to get into. Those are various custom models that have been created for Doritos, right? And so, and I love this, and you know, there's this aspect of you're using custom out models and you're going to start to generate very good creative outputs. But there's impact, right, before and after the creation of it. And so you're going to need these interfaces to help you derive that value. - But we'll get to that. - Yeah. And you are definitely the better expert at talking about it than me. So thank you for chiming in.
So how does this all come together? The world is big, as you all know. There are so many ingredients and so many different foods all over the world, so delicious.
But as you can tell, this is hard to do a global culinary platform and still make it locally relevant to all of these places. Also, shout out to Adobe Stock for all these photos.
But it would take a number of time and resources and oversight that we really don't have.
And as a person at the Design Center, we want to make sure we have design excellence and making sure our food looks proper. So luckily, we have custom models and the last sizzle reel, I promise, but it's a fun one.
- Yeah! - I'm getting hungry.
Is everybody going to go get Doritos after the-- That's the plan. That's the goal.
And I want to start off where we originally started. Does this look like design excellence to you? Yeah. This is a whole audit of what we pulled. It was from chefs from away from home partners. And we were like, we cannot let this out here. But what if we can take every ingredient on this page and have a magic button and turn from this into this? This is my "Ooh, ah," page.
But I promise you this was not just a magic button.
It took a lot of handholding with Yuji and the team. And they were super helpful with all of our questions we had with custom models because this is a brand-new capability.
But this is kind of what we've been working with. And not only that, with all the touchpoints from the photo shoot I showed you before, we can also do single chips. So a lot of flexibility with custom models.
All to say that we are unlocking high quality for personalization at scale. Are you sick of seeing personalization at scale yet? This hopefully, will be the last one before you go home.
And excited. A lot of things are kind of in the process right now because we launched back in October and we released this to markets in January. So to get this in time I had to bring some things together, but our first pilot market was Turkey. As you can see, these are all the custom model outputs with some help from our freelancer.
And they're actually doing this on bagel carts in Turkey. So it kind of creates a new revenue for local food trucks where they have bagels in the morning and then they can create this food truck at a later time and have more recipes for people.
So to say that, I mean, we have Turkey, but I think I wanted to show you some of the key takeaways and what we've really seen through custom models and how we can make it super personalized. But so Doritos Loaded is in 30 markets, and all of them are tailored differently. So we have India that's doing kiosks in shopping centers. We have the UK that's doing it on pack. We have Taiwan. We have so many different markets and they're all very different. So we want to make sure these recipes are tailored to them. Away from home partnerships. So as you know, PepsiCo, we have partnerships with KFC, Pizza Hut.
Who else? Subway. So and all of their menus look very different. And we want to sell in Doritos Loaded. So when we have this, we can create the recipes based on their menu and then they can even leverage those recipes to put on their menu. And then it's all in the triangle and we're all happy about it.
Content support. So PepsiCo has a new thing called content kitchens because culinary is the in thing now. And we work with influencers, chefs, and I think Tina was the last one, Tini was the last one. But just to have content support to even show these influencers, hey, we'd really like our food to look like this. Can you do your thing? And also, kind of give them a reference.
And then out of home seasonal and trend-based, you know how fast trends come and go. And with Firefly custom models, we can do this so easily and be on brand and share it out with the world.
Also seasonal. So we have some people wanting to generate stuff for Christmas, Halloween, possibly TV shows that I cannot tell you about right now, but hopefully, you will see it out in the world soon.
But all to say with the magic button...
We still need this process and we still need this human power. And even with aligning with the graph that I showed you before, we're trying to keep humans at the center of this. And we're not replacing humans, we're just making supercharging all the capabilities that they do. And even the photoshoots we have, we're adding the most value to them. So we still need the art direction. We still need the photoshoot, we even need all the retouching afterwards because we want to make sure our food looks great. But it's that red model part where we're really amplifying everything. So I hope that everybody found this useful. And if you did, Yuji's going to talk to you a little bit about best practices with Firefly custom models. Before I go there, I just want to add in. I know we're talking about custom models, right? I know we're talking about innovation, about creation of content at scale. I just want to throw in because I had a lot of discussions today. I talked to a lot of folks, and there was one overarching theme. If we unlock this, if we unlock the friction when it comes to content creation, really there's a possibility that we're transferring the friction. You think about all of the steps before content creation and all of the steps in terms of the activation. So keep that in mind as you're unlocking content creation that there's other aspects that you're going to go to. And so that was amazing what Veronica showed us, and I've seen it many times. I continue to be amazed and she hasn't really-- Yeah. She hasn't really talked about-- Thank you. Thank you. Thank you.
She hasn't said this, but she got her promotion, right? - Oh, gosh. - She's been recognized. I'm blushing.
I wasn't expecting that. Thank you. She's got a dog. And then she-- I know, so much going on right now, but I'm here in Vegas.
I'm trying to say she's a rising star. We feel privileged to partner with Veronica. Thank you. But when I think about all of the great things that Veronica and team have done, I start to think about everyone else, like all of you folks in the room. How do we unlock that? How do we unlock custom models to create great innovation? So that's what I want to talk to you guys about today. I want to talk to you about best practice and how we can get there.
So the first one that I want to talk about, so full disclosure, my notes got-- Oh, I think some of them are here now. Oh. Oh, it crashed and it came back to life. That's first for technology. So we're talking about optimizing your training data set, right? And so, first and foremost, it's all about consistency. We really want to create that uniform style. And so, let's unpack that a little bit, like, what am I talking about. Take a look at the images that we've put up. What do you see? I see different lighting. I see aspect ratio, I see the background, I see the foreground, I see the composition. All of these make up sort of the style of these images. And so we want to make sure that we have consistency in the style in order to drive really great outputs.
So you know, I just said we want the consistency. We also want variations. Sounds like a little bit of conflict. Let me dig into that a little bit. It's not. So maybe when we think about embracing variations, I might think about a slightly different lighting...
Maybe a different angle. These are all forms of variation. You want to create examples, different examples within certain styles to have those variations. So let me make this into an example to really bring it home. So if you look at the models in these pictures, what do you see? I see a male, I see a female, right? And they're all wearing different outfits. So this is what I mean about embracing variations but maintaining consistency.
So your first pro tip today, right, is to curate your data set thoughtfully.
Think about the consistency, and embrace that variation, right? That's going to lead you to your good outputs. Oh, you know what...
Let me ask the room. Who here has heard the term auto fitting? Oh, sorry, overfitting.
Raise of hands. Overfitting. So I see a few. It's more than me. So it's a GenAI term, right? It's talking about the model becoming too specialized. It's becoming too narrow. And this happens when you're not following those two tips, right, about consistency and embracing variations. It starts to produce bad results. It's unable to create the great outputs. And so your pro tip here is to have that consistency and embrace the variations, and that would lead you to avoid the overfitting problem. And I asked for this because this will come up a few different times.
So the next thing that I want to talk about is aspect ratio. So we already talked a little bit about aspect ratios. It's one of the elements of the consistency that we talked about. You know, I bring this up because I think it deserves a special call out. When you think about your training data set and you think about the outputs of your generated images, there needs to be a correlation, right? So let me break that down. So if you think about-- If all of your images are going to be in portrait mode, your training data set should reflect that. And then, so what happens when you don't consider this, right? So if you think of your portrait mode and then you're generating a square output, something's going to get cropped out. You're going to lose something, right? And ultimately, you're trying to establish that consistency to drive the better quality. But if it's square it's no longer at that consistent image, right? So if you intended to have those square outputs with the portrait outputs, it should be reflected in your training dataset.
So your pro tip here, plan ahead.
Think about your aspect ratio early in the process.
Align it to your applications output specifications, right? This is going to save you time and it's going to save you energy of fixing stuff because you didn't realize that square was coming.
This is an important one. I kind of thought maybe put it up top. I didn't. It's important because it's overlooked, right? So what do I mean by crafting a flexible caption format? I don't know how much, how many of you are pretty new to this. Let me describe the Firefly custom model process. So you create your custom model, right? And then you've carefully curated your images that you want to train on, and you uploaded. There is a caption service that would describe those images for you.
And so what you want to do is you want to avoid a fixed format. So what I mean by a fixed format, right? So for example, the caption could start or end with the same phrase, right? You know, I always tell my daughter, "Use your words," right? There's many different ways to say something. So embrace the variety in how you describe that particular image.
Encourage that variability. So with those different captions and different styles, you can have different tones. You can make it short, you can make it long. You can say it in many different ways. So you want to embrace that.
Was there a pro tip? Oh, yeah, sorry. Your pro tip, test, test your models, right? So test it with variety of prompts so you can see how well your outputs are.
So similarly to your format, your fixed formats avoiding fixed formats, there's eliminating fixed phrases, right? So if something's repetitive it might start with, this background is black, this background is black, this background is black. So what this-- What you're doing here is you're creating a disproportionate importance on a certain phrase, right? And that influences your model and would influence its results as well. So you want to make sure that you're reviewing. You know, don't take all of those captions and make sure-- Take a look at it. You know, first you've made sure it's not this fixed format, then you need to start looking at those repetitive phrases. And you start-- You should remove those.
Yeah. So let's take a look at this. So you can see, you can just see by visualizing that it's got a variety of different-- It's got short, it's got long, and you know, we've removed sort of those fixed phrases.
Okay. So ensuring-- Wait, did I miss one? Yes, I missed one.
Refining captions to capture core concepts. This is my favorite.
I'm a huge user of AI for meeting notes. You guys too? Yeah. We are. And I always, when I use AI to create my meeting notes, I always check it, right? I always check it before I send it out. Every now and then, it just captures the wrong thing, right? So that's what we're talking about here, does your caption align to the core concept that you're trying to create? We know that when you created your model you carefully crafted the images. You had something in mind when you're creating that. And then when you did your variations, you had something in mind. You were purposeful in training your data set. And so we want to make sure, just don't take what the caption service gives to you. Go through it, read it, right? Pick up the nuances because those caption services isn't as bright as you are. It's not going to pick up those nuances. So just don't accept it blindly. If you are the custom model designer, it's on you to make sure you're going through and you're making manual edits to make sure that that caption aligns the essence to your core concept.
So your pro tip...
Don't just take the caption services, right? Go through and manually edit to align to your core concept.
Okay. So the last one, we've come to the last pro tip. I know it's Thursday morning. You guys are tired. You guys are doing great.
But I'm going to change this one up. So I'm going to make this a test.
You guys can't leave until you figure this one out. Take a look at the image.
Tell me what's wrong with it.
Tell me what's wrong with it. Anyone? Yeah. [Man] I see pizza everywhere, but that's not pizza. - Yeah. - Oh, all right. He wins a Dorito. Oh, a Dorito.
There we go. - Cool Ranch? - Cool Ranch. You get a choice of them all. There you go. Exactly. It's not a pizza. The caption service, it missed it, right? So it's on you. Like you might think, "Yeah, it's got it." And so you stop looking at it. You have to really pay attention. You have to really proofread what's going on.
We're not making pizzas, we're making Doritos, we're making-- And so you want to go through. Ensure your caption for accuracy, right? And it matters because it-- This is critical, right? Accuracy is critical to your model's ability to generate great outputs.
So your pro tip here, incorporate checking those caption services, right? It's not just initially, like continually check to make sure that it's accurate and make those refinements as needed.
Great. Well, I mean, I've seen this and you've taught us this and we've worked so closely together. But seeing it again, it just is always so helpful, even just with building it and making this food look great. So thank you so much, Yuji. - You're welcome. - So yeah. Oh, if you want to go back to the other side. So I think we've seen such a big unlock and remember that circle that we showed you before. I think we're going back to the bold implementation and finding the why between all of our PepsiCo brands. You know, Doritos Loaded fit perfectly with Firefly custom models and we see the potential and we just want to keep that momentum going. So we're finding the why. We're also creating a pilot for our employees and doing a kind of education tailored experience kicking off actually tomorrow. So we'll keep you posted how that goes. And then also just getting feedback from them and trying to find the right projects and making sure that this, like, easily goes into their workflow. We don't want to add any more tasks, but we see all the unlock this can do and we just want to share it with the Design Center and everybody else.
So I guess that I hope Yuji did a great job doing key takeaways and I hope everything that I said to you, you can find your own path and use it within your companies. But here are just a few of the takeaways that we want to give you outside of just giving you Doritos bags.
So empowering your people, we want people to work smarter and even especially when resources are limited. And these tools can expand reach and influence. So we really want to make sure that our teams know how to use this and that we're enhancing them, not replacing them. Future proofing the team. I think as you all know with Summit and MAX and everything, AI is not going away.
And we want to make sure they know all these programs and even going beyond PepsiCo. When they want to go to their future career path, I think this will be great for them and then a good unlock and long-term success in general, not only for them but for PepsiCo as well.
Ethical AI. I think we all know a few other platforms out there that just pull from intellectual property and we do not want to do that and we don't want our intellectual property messed around with. So that is why we've partnered with Adobe because they have the ethic and responsible way of using AI. And we can show you some of their policies if they haven't talked to you already about that. And then also PepsiCo's as well. Work very closely with Andre to make sure that all of this is appropriate and for our company. And then maximizing impact, like I said. And I always love a good car pun. And with F1 last night, we're supercharging all the value that we have and we're still keeping in mind all of the creatives and we're just really amplifying that power. Yeah, I want to just double down on that. Go for it. Yeah. It's really about giving everyone superpowers. We want to democratize all of this. And so the end user experience for everyone is highly personalized experiences, and that's what all of this is about, and that's what we are trying to unlock. Yeah. Well, I think that's it for most of us. I know I think we have some time to take a few questions. If you have any more questions or want to keep in touch with me, I have my LinkedIn on here, and if you create any Doritos Loaded recipes at MAX. People sent me some, so please share them. I would love to see it. So well, this was amazing. This was a really good chat. Thank you so much. Yeah.
Thank you. [Music]
