Cutting Edge AI

#3 How Generative AI Is Transforming the Way Fashion Creates with Julius Harling (Graswald AI)

• Angel Invest • Season 1 • Episode 3

Share episode

Fashion imagery is entering a new phase, one where AI doesn’t just edit photos, it creates them. Graswald AI, founded by Julius Harling, builds a digital photo studio that lets brands generate lifelike product images and campaign visuals without physical shoots. Using proprietary generative models trained on synthetic data, Graswald can dress virtual avatars in real garments with pixel-accurate fidelity to fabric, texture, and fit.

In this episode, Julius explains how their multi-model system works: from training AI on structured garment data to creating consistent lighting and brand-specific aesthetics. We discuss how digital avatars are designed with real emotional expression, why brands are demanding exclusive AI models that reflect their casting choices, and how synthetic data pipelines enable scale without copyright risks.

The conversation also unpacks the economics behind this shift: Why photo production costs can drop by more than 90%, how AI lets creative teams test ideas that were previously too expensive to shoot, and what this means for supply chains, sustainability, and time-to-market.

We close on the bigger picture: the convergence of multimodal foundation models, falling inference costs, and the rise of context-aware AI systems that will make content generation as fluid as sketching an idea.

If the past decade digitized retail, this one is about digitizing creation itself.

Julius Harling: [00:00:00] We always see, especially fast fashion, as like these really evil big corporations that just want to press money out. But the people working at these companies are genuinely some of the most creative and good people I've met in my life.

[00:21] This is Cutting Edge AI, brought to you by Angel Invest. With your hosts, Jens Lipinski and Robin Harbour.

Robin Harbort: [00:32] Our guest today is Julius Harling, founder and CEO of Graswald AI. Julius is currently shaping the future of fashion product images. With a background in 3D design and a deep love for art and music, he now leads one of the most innovative AI visual companies. Back in August, we talked to him about his company, Graswald AI, and they are building the digital photo studio for fashion brands. This is the Cutting Edge AI Podcast by Angel Invest. Let's go. Hi, Julius. Hi, Jens. Great to have you here. Julius, how are you doing?

Julius Harling: [01:19] Really good. Really good. Can't complain. How about you?

Robin Harbort: [01:20] I'm fine. And Jens, how are you doing?

Jens Lapinski: [01:26] I'm good. Calling from London. Nice. Back after a few years. I haven't lived here since 2014, but it's always good to be back.

Robin Harbort: [01:34] Wonderful. Julius, really great to have you here today. You're building something in the creative industry. Maybe we start with a hard question. Do you think with all the AI stuff happening, is AI killing creativity?

Julius Harling: [01:50] No. No, not at all. Not at all. Also, not a very unique view on this topic, but in general, AI is a tool. Regardless if it's generative AI for video, images, or just large language models and like any other tool can help you do stuff. You know, I know this is like always the same thing to say about, oh, AI is killing jobs or not, is killing creativity or not. It's not, it's a tool and it is really good at, first of all, like replacing repetitive tasks. And at the same time, it is actually quite creative on its own. So one of the things that I think AI does do very well, actually, especially when it comes to Gen AI is creativity. And that's awesome. We see this on our end, like what we do essentially is we dress AI avatars with clothing. So the AI needs to be quite creative in a certain sense to figure out this is a flat shirt, how does it look on the person without changing the shirt itself, like imagining that person in a different environment. What we do see is that, talking just about my business and our use case, things that before took 100k to shoot, right? You know, you get your you book in the entire island, you get your whole crew, you know, on the plane and flights, and then then it's raining, right? And you're completely fucked. There's like nothing you can do. And then you have to extend two more days, but then the models has another shoot. So yeah, the model is leaving, what do you do? This obviously, you know, can be automated, or it can be streamlined, maybe not even automated much, much faster. And what this is allowing our clients to do is actually be bolder in what they're doing, right? I can pitch an idea that's completely crazy. I can pitch 10 ideas to the CEO and say, look, I have these great ideas. Like two years ago, the CEO would say, yeah, but we can't afford even testing this. It's way too expensive. Now it's like, okay, great. Let's do it. Let's try it.

Jens Lapinski: [03:46] You probably wouldn't even pitch an idea.

Julius Harling: [03:48] You would not. You would

Jens Lapinski: [03:49] Just show the results and say, I did this. It took one minute. What do you think?

Julius Harling: [03:53] Yeah. And yeah, we can go live tomorrow. Great.

Jens Lapinski: [03:56] Right. So just for the audience, maybe you explain what you do in the company, and then we can come back to how that impacts.

Julius Harling: [04:05] So at Graswald AI, we're building a digital photo studio for fashion brands, we're doing two things, essentially. So I mean, imagine any fashion brand you can think of right now, you go to their store, either physical store or retailer, or you go online, you see images of the models wearing products, and based on how they look, you will probably buy them or not. right. And these images are right now shot in studios, big brands shoot 30, 40, 50 products per day. It's very, very expensive. It's very time consuming, you have to, you know, send models back and forth, it's just e commerce. And then you have all the more like campaign stuff, whenever you see a person I don't know, enjoying a drink on a hill, wearing clothing or in a certain environment. Lifestyle imagery. That's even more expensive to shoot, but that sells more of the brand vibe and the story. And the reason why we essentially buy things is because we like the brand. So it's really important for brands. But it's very expensive. And we do two things. We help brands digitize this process first by completely digitizing and automating the whole e-commerce PDP content as a product description page content images. So essentially person wearing the outfit from the front, from the side, from the back, maybe a funny pose and maybe a video. And then we also help them use that input to create more creative lifestyle content, imagining that person wearing the same outfit in a different environment or with a different interaction, different action, and then doing this very, very consistently based on the brand's general visual guidelines and brand style and messaging.

Robin Harbort: [05:40] How do you make sure that the AI avatars wearing clothes from brands really look like the actual clothes and that there are no hallucinations and stuff like that?

Julius Harling: [05:53] I mean, we've trained a very good AI model that is just very, very good at doing that.

Robin Harbort: [05:57] How did you do that?

Julius Harling: [05:58] We've been working on this for a while now, and we've been starting with playing around with different technologies that are there, that are great at trying to get certain, you know, garments on top of, of a person, and there are different approaches to that, that have been around for quite some time, they all work, okay, they're very slow. And they're really, really cool to play around with that. But if you go to a brand, like, yeah, you know, imagine this, and then they have their high end shooting e commerce, that's, that's not good enough. So in the end, we, yeah, I mean, we trained our own custom model, what's, I think, more interesting is that we we've developed a system to generate, like we train the model to train a model to train a model to train a model or like, you know, something like this. So we developed a system that allows us to generate synthetic training data for very specific use cases in a very short amount of time at a very, very high quality. And that allows us to essentially build very, very robust, very accurate models, not rely on like external data, or we don't have to like I know, crawl and illegally obtain fashion images from some brand. And it also allows us to scale up really, really quickly if we have different product categories that we want to add. I don't know, footwear, accessories, bags. How do you do this? AI, especially when it comes to visual stuff, it's really sitting in front of images for a long time and figuring out what should go into your training set and what doesn't. And it's very, very, you know, Not a very fun task, but you need to be very good at curating these sets and coming up with very intelligent ideas on what should we, how do we, from one image to another, help the AI understand what it should do.

Jens Lapinski: [07:40] How much of that is effectively providing a certain type of context to the model? And how much of that is actually providing data? 

Julius Harling: [07:52] How much of the training process or? 

Jens Lapinski: [07:55] Yeah, because if you think about it, so they let's just assume that the models will become both more powerful and cheaper over time. So the cost to run them will drop quite a lot. I mean, it's been dropping, how much Robin per year 80% 90% something like that.

Robin Harbort: [08:06] 80% per year

Jens Lapinski: [08:07] And then at the same time, the performance has gone up because it become intrinsically better. So with the same amount of money, you can now bring far more power to bear, and that then improves practically everything. So that's one aspect of it. And the other one, so that basically means if it continues, if the costs continue to drop at this rate for the next five years or so, running a model cost effectively or at very, very high power, shall we just say, becomes incredibly cheap. Yeah. So then the question is, how much training, let's just say in five years time, how much training do the models actually need in order to be able to do what you are doing? And how much number one, number two, but then there is this other element, which is then saying, hey, look, here is this style of this brand. This is how these garments, what they are made of, this is what that then actually looks like. So there's contextual information that is both about the garment and the brand that's making it, and certain type of style guide that they have that no model will ever know unless you tell it to the model. So there is an element. I guess the question that I have is, what element of what you do right now is actually connecting models with the brands? And to what extent do you do you still think that you need to tune the models to make them do what you want them to do?

Julius Harling: [09:30] Very, very interesting question. Let me break that down, first of all, into like what we're already doing, and then where I think the like, no one, no one knows where we'll be in five years, if we're even alive. I don't know, like, No knows what's going to happen in two years. But in general, I mean, I agree, how we're doing this right now is besides the models that we're doing, we're doing the way that we've built the technology, which is not just, you know, one AI model that handles everything. But it's a specific setup that allows us to have very strong consistency by having certain models that are good at certain things and doing certain tasks. And that allows us to fine-tune certain things very well for the brand. For example, how does the avatar look? How is it styled? How old is the avatar? What does on-brand mean? It doesn't just mean how the how the garment is scalded, it means you know, it starts with the casting of the the avatars, the lighting that's being used. And this is also something that brands continuously iterate. I had a call with with one of our clients like last week, and we were talking about some lighting adjustments that we did for them. Like, you know, what do you think? I was like, yeah, that's great. But ignore this other thing that we share with you, because that's an outdated lighting. It's like, okay, our data line, like, how often do you change it? It's like, ah, not that often, like every three months. And it was like, so in general, I think, I mean, again, being able to change these things more, more quickly, and be more creative, this this touches on the first question is one of the one of the strong benefits. Like, I honestly do believe that we will get to a point and this probably not what what everyone wants to hear. But like, we will get to a point where models become like foundational models, essentially, like, for example, omni models that are multimodal will become so strong, that there is very, very little specific, you know, niche value, the way we currently think about value, and currently think about products that can be done, I don't need to go to a product, if I can just say, build me this solution or think of this solution to solve the problem for me to one interface very, very easily and quickly and at the scale that I want. What we're seeing already with large language models is models do need less and less context. to do deliver more and more high quality results like the way I interact with, for example, I don't know the latest strategy reasoning models completely different to how I interact with them a year ago now I just you know, I would never want to talk to my team that way because it's like almost you know, like, I can't imagine a good output if I just throw some shit at the team member. And where's the context? What do I actually have to do with with the latest models, it's already really good at getting the context, even if I provide very little and giving me the exact output, I think this will, this will increase. And I would love to be the one that says, I know how things are in three years. And what we're doing right now, and what we're building right now will be great. In three years. I don't think that's that's, I think that's magic. That's not like, like something one can actually do. What we're seeing right now and why we are right now succeeding is that we're really quick at adapting. We look at what's happening in the market and we are the fastest to innovate. We're the fastest to adapt to new technologies and we kind of make the smartest, most intelligent choices on how to combine them to solve the customer problems. At some point, you know, we might all think about, okay, what problems are still there to be solved, which I think is a very interesting question, probably off this conversation as well. But I think this is not a question specifically to image generation or what we're doing, but foundational question we'll all. have to ask at some point. What makes a great person model and not an AI model? I didn't really think about that before I started working on this. I thought it's like, you know, you need to be pretty and like have the right body type and etc. One of the main things I've learned is that it's really a lot about like, performance and like the expression and the energy you send. So we you know, one of the reasons why people want to work with AI models is because they can never get tired. Usually after a shoot, full day of shooting, the last three, four outfits, they're not as high quality anymore because the model gets tired, the eyes get tired, the expression is just less present. And usually the models that perform very well, the actual real world models, apparently are the ones that can perform, that have energy, that you connect with. And that's what we are putting an incredible amount of time to do with the virtual models as well, making sure that they not only look like real humans, but that they really have character. We generate these models together with our clients most of the time exclusively. So, you know, every brand wants their own AI models. When we like have to accidentally like smile, when we see like a person looking at us, then we know like, okay, that that was right, right? Because it's a natural reaction that we get from, you know, seeing another emotional reaction of a person, even if that person is not not real, like that, that connection, that that needs to be still still real, regardless of, you know, if the pixels behind and show a person that exists or doesn't.

Robin Harbort: [14:29] If I take a look on your LinkedIn profile, there are many of those really realistic AI avatars. And what I really noticed was that you give your AI avatars names. Is this also part of the, let's say, strategy to make the AI avatars more human or establish a kind of emotional connection?

Julius Harling: [14:53] Honestly, not really. When we started doing this, we were just, like, hated referring to them in any other way than giving them names, because it's like, okay, which avatar are you going to use? DX3-750? Like, no one can work with that. Right? So by just giving them names, it just made it simpler to work with both for us and our clients. But it does have a nice taste, right? Like it is, I think, and we do, we do to some degree, like we still play with that, with that fact that these images don't depict a real human, right? They don't exist. But the connection that you feel to them is probably still real. And if you look at a photo shoot, the person's also not real. They've been retouched for hours. Right. They've been styled in, you know, very, very lengthy processes. Like the person behind that does not at all look like that. And yeah, we want to have like more authentic and more real images. The amount of work that goes into making an image authentic and real is insane. 

Rovin Harbort: [15:52]Julius, what is here your personal opinion on declaring AI content as AI? Do you think we should do it or will it be obsolete in two years anyway, because nobody will care if it's AI or real?

Julius Harling: [16:03] Yeah. Like this is really just my personal opinion. It doesn't reflect our company's policy. I mean, first of all, I think brands should do what they want and what they feel comfortable with, and these are decisions that we're seeing that brands do differently, right? Some brands are like authenticity and being like very direct about our use of AI is really important to our brands and other like, no, you know, or just, let's see how it works. Let's just push it up. I honestly think it doesn't make a huge difference. Like in the end, I will make a purchase if I connect with the identity of the brand and if I like the idea of me wearing that clothing and if I can relate to that person wearing that clothing in a way that kind of pushes me to make that decision, I don't need to like, like really, if that person exists exactly the same way that it looks there or not does not have any impact on my personal bio decision. I think it does not for most of us, I could say, Oh, but my values are like this and that. And I don't like kind of models, you know, being replaced. So I don't want to purchase from brands that use AI models. And you know, then it might be helpful for your buys buying decision if there's a indicator, but you wouldn't notice. And we did a test with a client that I cannot name, but in the end, they did a survey, they had like images of us and images of them. And they asked consumers, would you prefer AI models or real people? And which of the images do you like most? And almost all of them said, I would prefer real people. And almost all of them actually preferred the AI images. 

Jens Lapinski: [17:33] It's just fascinating, right? It's just, I mean, it's effective, you're just shooting the breeze. And it's not really about technology. It's more about sociology, if you like. But to say these AI supermodels. Yeah. So they will, because there will be some that will just be immensely popular for, I don't know, it's just the eyes little bit like this. And then, you know, whatever has some imperfections, but they're really charming and love it and whatever. Right. And then they just say that it's owned by, well, I don't know, some brand has got a proprietary right to it. Yeah. And then suddenly somebody comes says, I want to license that model. I want that to be the star in my next movie or the other way around. They make a digital movie and I say, Oh, we're going to want that want to be our blah blah, you know, something like that, or computer game. And I saw it's all becoming completely much more interlinked. Yeah, with a photo shoot, you're putting you don't really know these models, because they're just you just experienced them visually. But you could drive them in a computer game, they could be in the movie, and it could all be they could have a certain type of personality. Yeah. Have you seen any of that starting to emerge? Is that starting to happen somewhere?

Julius Harling: [18:38] Yes, definitely. I think there is a difference though, on something I learned during this process. I mean, this is the situation you're describing is already happening with influencers, right? And essentially, they're real people, they're not AI people, but it's exactly this. You have the influencer and they have their own, they work with, I don't know, About You and they have their own clothing brand that's that's on from About You, but it's actually their own brand, Teddy Tecla brand, for example, or whatever. And they, you know, are, you know, having a different job, but then they also have the clothing brand and this and this and that. And then they're part of the video game or that movie. These are influencers, but they're the exception. Like you don't have a brand with, I don't know, like every single shirt is worn by that influencer because then it is not your brand. And I think that's the same with AI. And we're seeing like there, there's, I think around Wimbledon, the tournament, there was one AI character that apparently was there and posted like images on Instagram and LinkedIn and people got crazy and it's like, oh, this new AI influencers keeping everyone, I don't know, like excited. Obviously she wasn't there because she doesn't exist. It is working and we're getting to that and I think we will see this, but I think we
will also see a certain fatigue. Like what we're getting right now is excitement because it's new. In the end, with everyone wants to become an influencer to some degree and then, you know, very few people find an angle that is so unique as a brand that makes them stick. And I think it's the same thing. Like just because I can have a pretty face or in an interesting facial, you know, structure, it doesn't mean that I will be successful. It's all about building a brand that's, by the way, really interesting. Fashion companies are the only companies that we don't call companies. We call them brands because it's really not about selling, selling
a t-shirt. Everyone can do a white t-shirt. You really just buy into the brand. And that's again, is it klling creativity? No, it's actually enforcing like the only people will have a good job are the people are really creative, really good storytellers who can make you stand out, who can build really good brands. And then it doesn't matter if you do it for an influencer face or a fashion brand or whatever. Branding is everything I think in the space when you can create whatever you want. How do you stay unique and relevant? 

Jens Lapinski: [20:50] What's the most surprising thing you've learned
in the last year about the space? 

Julius Harling: [20:57] I would have answered differently without you saying about
the space. 

Jens Lapinski: [21:00] Well, give two answers. 

Julius Harling: [21:02] I start with about the space and then I'll let the other stuff.
Honestly, the sounds really dumb because it makes sense when you think about it, but I had no idea how multifaceted this space is. It makes such a huge difference of like there's so many different types of brands and so many different types of styles and they all work in their own very unique specific social circles. And when you sell to them, when you interact with them, they all value different things. I can talk to a, you know, someone who says I'm a fashion photographer, but it makes a huge difference if they're shooting e-commerce content for Zalando, or if they're shooting campaigns for a million for Louis Vuitton, right? And then within that,
and even like being a model, completely different jobs, completely different lifestyles as well. It really is insane. One thing that really surprised me is we always see, especially fast fashion as like these really evil big corporations that just want to press money out of the world. And they are doing a lot of bad shit for the planet. It's not very good. It's not very sustainable. But the people working at these companies are genuinely some of the most creative and good people I've met in my life. And they really, really care about a story that has meaning and like doing something that is artistically on such a high, high level that I think is very, very impressive. Did not at all expect that. I thought it was very, it would be much more commercial. And I think that in general, what I learned about this last six months,
which I think a lot of founders might resonate with that really surprised me
is that how much can be achieved or how much I as a, as a person, as a founder
can, can achieve if I just, you know, when I'm not afraid and I don't think about,
Oh, can this be done? But if I just very strongly believe, yeah, I can do it, we can do it. It's just a matter of how I've been really surprised. Like when I started slowly shifting to realizing, yeah, we can do things that we thought were impossible just two months ago, and we would not ever be able to do that, especially with the resources we have that I think is the most impactful thing I've learned in the last six months.

Jens Lapinski: [23:10] The goals that you set for yourself are always positive for you. Goals are inherently positive and positive goals make people optimistic
and they bring joy to, to everything. So if you want to have a joyful life, set yourself goals and try to hit them. Because that gives you a positive outlook on life. And it's one of the things when you say, oh, everything, politics is so pessimistic or whatever, everything that's pessimistic. Ask yourself, what goals do these people have? Do they even have any? Or do they communicate any? Does our society have goals? I don't know. I mean, I haven't heard any from any German politician in quite a while. You know, it's sort of like, I have no clue what they want. And it's sort of, I think that's the real challenge is defining beautiful goals for yourself and for your company and for the customers, because if you do that and you can reach them, it's just both the prospect of getting there and then accomplishing that is just fantastic. It's a super powerful tool. And then you're not afraid of going for them, you know, because if you say there's of goals and then you, that's where the burnout is. When you think you can't reach your goals, the desperation sort of burns you out. That's a negative spiral. That's what you don't want.

Robin Harbort: [24:33] I think we have to ask now Julius, what is the one positive goal? What's the main goal you currently want to achieve?

Julius Harling: [24:39] In general, I think everything I have been doing to some degree all my life, especially in the last years is building something with the aim of building something that touches people. I've been working on creative tooling all my life, meaning combining artistic processes with technologies, enabling people to do more with telling their story in a better way. Or especially now, where I thought when I started thinking about this, "Oh, this is so far away from where I started. It's not at all. Like we're still doing the same thing. And we, it's more important than ever that whatever people create with our tool creates an emotional connection to, you know, the output to the people who look at it and creates feelings for them. When we do that, we enable people to like, they generate an image. It doesn't matter if it's, you know, however it was created and someone looks at it and they feel something, then we've done our job right. And we've enabled the person who uses our tool to do their job very, very well. And regardless of where we go and what I will be doing, I think we'll always be around that because that's incredibly fulfilling. 

Robin Harbort: [25:47] And there's only one question left for me and you can only answer with yes or no. Will AI kill us all?

Julius Harling: [25:55] Yes.

Robin Harbort: [25:57] When?

Julius Harling: [25:58] I don't know. Funnily enough, I'm very pessimistic about the future, but I'm very optimistic about what I'm doing and what we're building.
And for some reason that works out very, very well for me. Like bad news or things I would consider bad news don't at all affect me. 'Cause I imagine it to be exactly like that or worse.

Jens Lapinski: [25:18] The way to think about that is that,
if they come for us, at least they will look beautiful because Graswald will make it so. That's the upside here.

Julius Harling: [26:30] Yeah, I'll be working my ass off to make sure that at least it will be a nice sight.

Jens Lapinski: [26:42] Okay. Well, one thing that we've noticed a lot is that the speed with which companies can move when they use AI tools increases a lot.
I mean, I have got founders who by using certain types of tools can produce two to three times more output in a certain period of time than they were able to do before, which means that the speed with which the company progresses, product
progresses, and the size of the team that's necessary, as it is with a much smaller team, you can move much, much faster and can get much, much, much, much more done, much, much more quickly. And I was just wondering how that is impacting the creative cycle, if you like. Because before, if everything is sort of analog, we need to pre-plan it and pre-decide this creativity now much more reactive towards how does like these AI tools change how we work with the new. 

Julius Harling: [27:38] You know, creative processes usually have two phases. There's the actual creative phase and then there's a production phase, right. Anything that's artistic, it has these two at a certain level of scale and
professionality and quality, it always has these two things. You know, you start with the creative process and you think, oh, what will be what would have an impact, blah, blah, blah. You do a lot of different things all over the place. And then, you know, whatever comes out of this, that's the thing that we want to do. Okay, but how do we do this really hard? We need this, this, this, this, this. And then this where producers come in, right? And try and, okay, that's a creative vision.
How can we do this? And they create very strong production plans
and we need to buy this, we need to get that, blah, blah, blah, blah, blah. One of the biggest issues of this process is that creativity is nothing you can plan. And it's also beauty of it, right? You cannot plan creativity. And some of the most beautiful pieces of art, regardless of are we talking about movies, music, whatever, they happen in the moment, like unscripted things in movies or certain lighting that was actually, you know, it was supposed to, the sun should be shining, it's raining, it completely changed the scene, but the director was like, oh wow, yeah, that's great. These things you cannot plan. And production, while it's an enabler of the creative vision, it's killing creativity because, you know, you need to follow a production schedule 'cause otherwise you're over time and you lose budget. What we're seeing now is that because we're able to create at a much higher quality level with much less processes and workflows involved, we can shorten the production cycle to a point where, ideally, at some point, you don't need it at all because production suggests a vehicle to get your creative vision to the level of quality that you need. But if your creative process already every single time outputs production grade results, you don't need it anymore. But that means you can focus entirely on the beauty of the creative process, which is, randomness, accidents, whatever is, you know, you didn't plan for whatever comes out and you get your result instantly. In some cases we are completely taking away, yeah, like the restrictions of creativity and I think what that will lead to is a much much much higher bar of quality in terms of what we actually consume and so the level of quality in terms of stories and art I think is going to just increase. 

Jens Lapinski: [29:49] If you think this further, right, it sounds to me that particularly in your space, it could really impact product development. Quite a lot because you can actually if you can do all of that, and you can change not just the how items are displayed, but also what the items are. Yeah. So then suddenly, you could say, well, hang on, make me 17 variations of this, and I want to play with that. And then basically, and then you suddenly think, shit, I need to make those shoes. Yeah, because that looks great. Yeah, that would really fit with this thing over there. Right? Has that already happened?

Julius Harling: [30:22] Yeah, that's, that's happening.
And we're seeing this already. Um, there are companies that we're also partnering with that do this. And I think that's something that we at some point, like some of our clients already kind of misuse our tool to do exactly that because it's possible. And, and we will add like from a business point of view, obviously we want to expand, expand around the value creation chain and we want to start at nothing. and we want to add the final product. We're starting kind of in the middle because that's the biggest pain point right now. We'll probably expand, but we're seeing this. There's even like really cool concepts and ideas. There's a company, I forgot the name, but what they're essentially doing is you can generate an image of your outfit with AI, and then they produce it for you specifically. So you have your own, you know, it doesn't scale very well, I think, but it's a really, really cool idea. And I think to some degree, this is where we're getting into, in any case, AI also means hyper-personalization. And I think in that clothing fashion thing, it's going to be even more so. And it's going to be even harder for brands than to keep their brand identity. Because if I'm personalized everything, like where am I? Where's the thing that makes me unique? Because it's all, you know, Oh, that's not, I don't know Adidas, because every Adidas looks different now, similar with like the iPhone home screens, right? Is that, does it sell Apple, say Apple? I don't know, because, you know, yours look completely different than mine. 


Jens Lapinski: [31:37] The most creative thing I've seen on Netflix in years is these short movies," Love, Death and Robots" I think it's called, right? And I think with new technology, you can probably make them at home if you wanted to. So then the only limit to how good it is, is literally creativity, storytelling ability, how you compose these things, you know, and how you make them up. So it's certainly, it's a great enabler, or it's a great leveler, if you like, right? You don't need to be big and powerful or anything like that, or have a lot of money or work for a big brand. In that case, you can be by yourself, have a lot of, you know, create beauty.
I think that's actually really, that's really cool.

Julius Harling: [32:21] We're still at this point where, and this I think a very interesting question that I also don't have an answer to like, at which point will we really just need creativity, right? Right now, there's still a lot of craftsmanship involved. And there is, in order to tell a good story, you need a lot of craftsmanship, right? I can't just say I have this idea of a story and that's my creative, whatever direction, and now create a script that does that really, really well. It will still probably not be as good as if a really, really well-versed copywriter writes that or whatever the right terminology is. And we're seeing obviously the more generative or the more free the output, the expected output is, the more AI can ship in and the more it's about nuances, the harder it is at the moment to get really the right context. But this will, I 100% agree with you, and it's like this will change drastically in the next months, maybe even. So at what point is it really, where does creativity start and where does it end? At what point do we not need craftsmanship anymore? And then how much of craftsmanship actually as part of your own creativity. And I just give this random idea, and then it's all what the AI does, but then it's not me anymore, and it's not that unique story I want to tell.

Jens Lapinski: [33:37] The craftsmanship will change. There is a very strong argument that you say that the real understanding out of what you do
comes because you're doing it, you're actually working it. You're really working on it. You are crafting it, and you get the feedback from both working with it, sometimes haptically, sometimes semantically or whatever, but the person is looking at it or consuming it, the customer and so forth, the viewer. And I think that will always be there by necessity. It's just that the feedback loop would probably be different, right? So if I'm playing with the AI engine and it's making things for me, then I have a certain, I'm crafting that. I'm not standing at the set and crafting it there, but it's just, I'm still crafting it. I'm still crafting. So I had a discussion with somebody on Saturday and they asked me, how do you think about that? And I said, what should we tell our kids? What should they do? Tell them the same thing you've always told them, go towards the future and play with the newest technologies that are out there and then lean into them irrespective of how they evolve or what they are. So in the past, so maybe learn to code. Now you say, well, you speak English, so job done. You have arrived at level three, so now English is okay. You know. Final word: What is the most cutting edge thing that you are doing right now?

Julius Harling: [35:00] Honestly, a week ago, we have released our newest model GAI3 that I think is like two, three months ahead of everyone else, at least according to what our clients and leads and prospects are saying. It's really, really, really good. It requires very-- like, no user interaction anymore. You just pick the products and you generate the image. It's really, really good. And it's really good at understanding the context of very little context, like a ghost mannequin image. It really gets the fit correct very well, which is really hard. Like, how should you know? And it's very fast as well. It's a lot of fun to work with. I just had our first live demo this morning, where I demoed it to the public, essentially. It's really cool. I get amazed at this every single day when I'm using it. And this enables a lot of other different things that we'll push in the next couple of weeks and months. But yeah, that's really exciting.

Robin Harbort: [35:56] Wonderful. Julius, Jens, thank you a lot. And hopefully you speak to you in the second part if AI doesn't kill us.

Julius Harling: [36:02] Sounds great. Thank you so, so much.

Jens Lapinski: [36:05] See you guys. Bye.

Robin Harbort: [36:10] Thanks for being here. If you enjoyed this episode, support us by leaving a follow and share the Cutting Edge AI podcast. See you next time.