[Music] [Zeno Koenigs] Hi, everyone. My name is Zeno Koenigs. I'm a Creative Technologist here at Accenture Song of the Netherlands. Accenture Song is the largest tech-powered creative group in the world. We have multiple offices around the world, and it's a blast to work here.
I am also the GenAI lead for Design & Digital Products here in Amsterdam, and I've been working with GenAI and Firefly, on delivering a couple of really cool projects in the past couple of months. We roll out about several different transcreations for more than 30 countries every week, and it's a lot of fun. So I'm going to talk to you today about some basics in transcreation.
And to do that, I have come up with a fictional brief, something that might be similar to something that we might encounter in our day-to-day. And it's going to be about pastries because I have a sweet tooth for the Instagram channels that I see where they make these very delicious looking pastries. And I don't know, it's very inspiring to see that. And then maybe one day, I can actually start noting or kneading some dough. The brief that we're going to use today is by a pastry company, and they have their base images that they use for their advertising and their social posts. And what's, of course, happening now is that you might want to create one which works for Halloween. And then so you go out, you get a studio and recreate that image with the fictional characters from Halloween. Now what AI empowers us to do at the moment is take that base image that they already have or in this case, we're actually going to create that together with Firefly. But take that base image and then transcreate this with the fictional characters from Halloween. So we can pick from anyone that we can come up with. This can be the grim reaper. It can be a witch or the mummy. And we're going to create that same setting using that same vibe of that base image that that company is already using for their official communications. So I don't know if any of you ever watched these pastry chef videos on Instagram, but they're always smiling and they're wearing white and they're standing in the front of this super nice white marble countertop. And what I want to do is I want to start with giving a bit of structure into the prompt by starting with the subject, which, in this case, is going to be the pastry chef. So we have a pastry chef...
wearing a culinary head cover...
and they're standing in front of this white marble countertop or behind actually.
And that sets the initial scene, right? So we have the subject and what's at the forefront of this image. And then we want to go into the environment where we say, "Okay, well, what does the environment look like?" Well, this is in an open space or open plan industrial kitchen, for example. And we are going to add the type of lighting, with bright even daylight.
And the cool thing about working with AI, I think, is always the fact that it's a little bit like painting, right? So as the late Bob Ross used to say, we don't make mistakes. We make happy accidents.
It's the same with generating images with AI because you're not doing it by yourself. I'm also an illustrator. When I start drawing, I know exactly that what that line I'm making is going to end up in. But now I'm working with something else, and I'm going to give it a prompt, or actually, an assignment. And it's going to translate what I've written into something else. And this can be inspirational. You can be happy with it, and you can get you going into more directions, "Oh, okay. Well, if I change this prompt a little bit, then maybe this is going to happen." Or you're not going to like it at all. You can start over, right? So we're going to start with this. So we're setting a scene, and it might not be directly what we want, but then we can change that quite easily.
So as soon as I hit Generate, it's going to come up with four different images based on that one prompt.
So it's going in the right direction, right? So we have our white marble countertop exactly like the one I see on social media, which is great.
Look a bit classic, and they have-- Well, they have the right headwear. I'm not absolutely fan of the baking trays here. So also, because it's social media, I'm going to change the aspect ratio to portrait so we have a bit more screen filling imagery here.
And let's say-- We're going to say it's professional kitchen.
You see, I didn't even write down kitchen to begin with.
And then we're going to say...
let's add a cake, right? It's a bit empty now here in front. And, of course, it's a pastry company selling cakes. So let's create-- Let's put in a cake...
with a one layer white base cake. So that's white frosting, butter frosting, I believe it's called.
And that'll give us a lot of space to actually go in different directions.
There we go. So this is a bit more like it. So as you can see, we have our one layer cake, and we want to have a large open plan. And what I like to do-- I'd like to add a camera lens to this. So shot on an 85 millimeter lens.
And what this does, it basically references the AI to look into its database and say, "Okay, what is usually shot with an 85 millimeter lens?" And an 85 millimeter lens is notoriously used for portrait photography and often with a very shallow depth of field. So this immediately already changes the type of images that the AI is going to look for in his brain, I'd like to say.
So what we can do next now we're getting this. And if we're not getting what we want, Firefly gives us these really nice little features where we can say, hey, you can select Bokeh effect, for example, that adds a bit of that shallow depth. But we want to use studio lights. As is for social media, we can select studio lights, and we can also display with the camera angle. But I feel like the camera angle is good here, so we're not going to do that. And I'm going to add select Generate.
And they're losing their head covers. So we're going to say, a friendly pastry chef.
Just to have space in there as well. So we can play around with this for a while and try to find that one image that we really want to use as the pastry company's base image. So I like this one, right? I feel like the hands are correct, the look and feel of the entire image, the open space behind him.
I'm not sure if we want to use this topping here. So what we can do next is we can-- Well, first of all, let's download this image.
This will give us a non-destructive way of working where keeping every iteration, we can always go back to the first version.
So what I'm going to do next is I'm going to go into Generative Fill, and it'll open up this window where we can start playing around with either adding or removing certain elements from this image.
By zooming in, you can see that the actual size of the brush stays the same. So you can get much more details. And if that is not enough, we can also go in, the brush size and hardness, as you might know from Photoshop, where you can actually go and change the size and width of the brush to get more detailed because it does have a maximum zoom in. As you can see, it's quite decent. So we can actually go and select this area of these beetroots that are on the cake, and we can remove them by just drawing on this image.
Because, again, these are not mistakes. They're happy accidents, right, as Bob Ross would say. We're painting over our beetroots. It's lovely beetroot pie.
Giving a...
turning this into a very easy-to-use base photographic photo for the pastry company that we're then going to use to transcreate something else.
Naming me a Halloween image because Halloween is coming up.
Well, we can work with this, and we can add white frosting to it.
And as you can see, I'm painting here. I also like to make the reference or the analogy towards painting itself when working with AI.
Because when illustrating or painting yourself, you might start with a blank canvas, and you have no idea in which direction you want to go. Sometimes you have something in your mind like, "Oh, I want to create an apple or a bridge." And you start going, and you start making all of these lines and sketches. And the further you go, then sometimes a shape or an image pops out, and you just tend to go with that. And in the end, you end up making something entirely different than what you had in mind in the first place. Now, of course, there's different types of designers and artists who do different things. But usually, for me, it works that way where I have something in my head, and I'm like, "Okay, I'm going to design something and create this or draw this thing that I have in mind." And then in the end, it might not be entire, I mean, it's still going to be nice, but it's not entirely what I had in mind in the first place. Sometimes these things change and shift. So I feel like that sometimes this is the same with AI.
You start out with a direction in your head, but you're doing it together with AI, right? So it's going to be different.
It's going to be different.
Okay, so we have something here, which I liked, it's a bit more base, and we can work with this. So let's keep this as our white frosting, and we're going to download this image just to keep it.
And then what we're going to do is we're going to upload this as a composition reference.
So what this does, it's going to tell the AI...
what the structure of the image is. And basically, it's going to give this coloring kit, right? So it knows that there's a shape standing there with its hands on the table and a cake in front. So that's basically the rule that it has now.
And we're going to have to imagine that this is the base image that's coming from the pastry company that we're doing this brief for, that we're doing this project for.
And so let's start out with creating a-- Let's say, the mummy, for example. So again, I'm going to start off with a mummy standing behind a countertop overlooking a, let's say, I think for the sake of the image, a red cake.
Because we're not only the director of this image that we're creating, but we're also having to come up with what is the-- Or envision, we have to envision this whole scene and then describe it as if you're a movie director, or as if you're an art director. So a mummy's standing behind the countertop overlooking a red cake with red eyes. I think that works.
And a smoky green...
background. And behind the countertop, I think what we're missing here, of course, is the rest of the scene. So in a...
horror sinister, in a sinister pastry shop, or kitchen, actually.
So again, now we're going to start with generating a new image but then using a reference image as base.
And we might want to start playing around with a prompt as we go along before finally hitting that final image that we want to use.
Now that's pretty cool, right? So we have this creepy guy or this one. I feel like they're all celebrating their birthday, though, which is cute.
That one's very scary.
And immediately, you see also that there's a stylistic change to this. So it taps into pop culture or like, what are the styles that Halloween creatures are usually created. But as you can see, it's following the stylistic or not the stylistic, but the composition reference quite well. So we still have our hands on the table. We have the hat wear is now like a turban wrapped around the mummy's head. And the red eyes are just very intense.
Well, it's Halloween, so it might actually be the mummy's birthday. But let's start adding or removing some elements to this as well because that is usually also one of the most fun things to work with. So I'm going to remove these flames because I feel like Halloween and flames or at least candles are a bit too festive. So I want to remove these and then they're gone. And then we get different types of gone, right? So we can see well, which ones look the least obstructive in a way. So which ones are the ones that you notice the least? So we have this. And then I saw something pop-up just now, which are little oat bars.
So I basically thought, "Oh, you're not happy with nothing here, so I'm going to give you something there." And that's not necessary, so we're going to select this image right here. So now we have a glowing cake.
We can actually go again and say, "Hey, look." Let's create the headless rider standing behind a counter...
pulling-- Well, with a...
pumpkin pie. Makes sense, right? That headless rider sometimes also wears a pumpkin for its head. And...
In a sinister kitchen...
with dramatic lighting. So let's see how that's going to turn out.
And again, we're using our initial reference image here on the side, and we can play around with the strength of it.
In my experience, middle strength is quite good when you want to go into a more creative type of direction. Because if we increase this into, let's say, very strong, you're going to see that it's probably going to try and also reference a bit more of the facial features of the person. And it might come up with a person standing here still. And we want to have that freedom of movement into these weird dramatics. Here as you can see now, we have these humanistic features that have come back. So we want to keep it at a more medium type of strength here.
And then we're back at this version. So Firefly also remembers that this was the previous one, so it just jumps back to our previous generation. But let's say the headless rider, I feel like what works well with orange is, of course, a purple apparel.
Wearing purple apparel.
So this is pretty neat. As you can see, it's very stylistic, and I like this, a bit more classical, fits with my image or my imagination of the headless rider.
Of course, as you know the reference image that we have has a hat, and it will try to keep that hat in. So what I want to do now is use Generative Fill, and I'm going to insert a Halloween pumpkin instead of this skull looking face. And I don't mind if it's a very big pumpkin head, actually.
I used to watch this TV show as a kid, and they had this pumpkin head guy. Not even sure what it's called, but it was a lot of fun.
Halloween pumpkin heads. Let's see where that gets us.
There we go. This is what I was going for. So this is quite nice. I like this guy. Let me just do a couple more so I'm sure that what I'm getting still fits, right? You can go endless. But in my experience, it's nice to make choices upfront quite fast. But, yeah, when you're doing this yourself, go ahead and make some more generations until you're truly happy. Once you've select Keep, right, you can always go in and make some more.
Yeah. Actually, the first one or this-- Nah.
I feel like this one is the best of the generations that we got right now. So I'm going to keep this, and I'm going to select Download. All right, well, that wraps it up for today. I want to thank you all very much for watching this. I hope it's all clear. This is, I think, the most we could do in about the time that we had, and I hope you enjoyed it. I did. And I hope everyone goes and creates something really cool. Thank you. [Music]
