Shallow Fakes, Models and Robots

Phillip: [00:01:18] Hello and welcome to Future Commerce, the podcast about cutting edge and next generation commerce. I'm Phillip.

Brian: [00:01:23] And I'm Brian.

Phillip: [00:01:24] And today we have what I'm going to say is my most anticipated interview of the entire year.

Phillip: [00:01:29] This is Yannis Konstantinidis, who is the founder and CEO of SuperPersonal. Say hello, Yannis.

Yannis: [00:01:36] Hi, guys. Very excited to be here. I love the podcast.

Phillip: [00:01:40] Oh, thank you very much. And I'm both a fan of and incredibly terrified of your technology.

Brian: [00:01:50] {laughter} Me too, man.

Phillip: [00:01:50] So for those of you who have not heard, you might want to go back and check out a couple of our prior podcasts. One we did live at Shoptalk about a month ago or so, and we talked a little bit about how deep fakes have gone commercial. And then we did one that was sort of covering the broader topic, the sort of existential topic of deep fakes, fake news, that sort of thing. And we now have real technology about a year later, which might change the way that you shop online. So, Yannis, why don't you tell us a little bit about SuperPersonal. Tell us about your background and what your technology does.

Yannis: [00:02:27] Yes, of course. SuperPersonal is a tech company. We create technology that enables personalization. And by personalization, I mean actually seeing yourself in the content. In terms of fashion, which is our initial focus, this means that, you know, it's seeing yourself wearing clothes that you've never won before. And our focus is to make the experience easy for the user, relatively fast without needing too much data and the result being flawless and photo real. Right? So you see yourself. You don't see someone that looks a bit like you. It's you. There's no 3D avatar that just looks almost right. So that was the initial sort of drive for us. How can we create this experience for the user? And my background is actually in production and animation production. I have a company which SuperPersonal is a part of. We have been doing it for the past 12 years. And we work with big companies around the world, big agencies. And yeah, two years ago we thought, okay, you know, how can we use this "new" exciting technology? Actually, it's not even there yet.

Phillip: [00:04:08] Yeah.

Yannis: [00:04:08] But how can we use this technology to create adaptive content? And that was the original brief we had in our minds. Meaning how can we you know, for example, we produce a video here, we spend months and months and thousands of people working and animating something. And then you have one static moving, a result which is consumed by hopefully thousands, millions of people. But then it's gone. You use it for 30 days for a campaign and then it's gone forever. So how can we take this new technology, this exciting technology, and create adaptive content? So content that changes over time or based on where you view it, or your mood, or whatever. So that was the initial sort of spark. And then we brought on board our partner here at SuperPersonal now, Jamil, who is an amazing AI engineer. He's the CTO of the company. And we started looking at different solutions. What can we build that can create this adaptive nature for the solution? So we looked at many different things. We started developing a few different things, and then we ended up in what we have now, which is an algorithm that is similar to, let's say, head swapping and face swapping technology. Only we wanted to make the user experience very quick, so that you don't need to provide too many data. The technology actually works with a 15 second video. In the app we have we asked for a bit longer, but it actually works with just 15 seconds. And then also processing time shouldn't be too long. And then the result, as I said, should be actually convincing. So you shouldn't be able to know that this is fake in any way. Yeah, it's why we're working on fashion now, because that's what we are focusing on, because there are more uses for it, obviously. So it's you can see yourself in films, you can see yourself in advertising and messaging. There are many, many use for this, but fashion has an obvious need for us. It always felt like, you know, people really need to see themselves wearing the clothes. The idea that we're using models to showcase clothing and outfits and so on is a bit of an antiquated idea the way we see it.

Brian: [00:07:11] Yes.

Yannis: [00:07:12] And we sort of started getting very excited about this idea, which you talked about you in that episode. You should go online, you should go to your favorite brand or place, let's say the SuperPersonal app or whatever it is, and then you see yourself. So you see yourself, you get your recommendations, but it's on you. You don't have to go and browse and find something that you're interested in and so on. So we love this idea. We love this vision. We think it's inevitable. It's going to happen. And, yeah, we're very excited to be working on it.

Brian: [00:07:57] Wow. {laughter}

Phillip: [00:07:59] Yeah wow. Wow is a great way of saying it. I think what people probably don't realize is they're encountering some form of this kind of technology every day, especially if you go to the movies. I just saw Captain Marvel and it's absolutely ridiculously uncanny that we can take Samuel L. Jackson, you know, in his late 60s, early 70s, and completely convincingly de-age him through film to to make him look like he's in his 30s or 40s. That's a shocking thing. But it's becoming commonplace, and it gets better every generation. There's some artistic work to that, right? There's some... They probably fill in a lot of the gaps. But would you characterize your technology as sort of an evolution or a starting point of what we might see in film already?

Yannis: [00:09:02] Yes, in many...

Phillip: [00:09:06] I'm OK for you to tell me no.

Yannis: [00:09:08] It is true, and like in many ways it's very, very true because there is nothing new here. I mean, it's actually this has been going on in cinema for a very long time and now takes two hundred thousand dollars to make it happen. I mean, up to now, and it is going to get cheaper and cheaper and cheaper and with this technology.

Phillip: [00:09:34] Sure. Sure.

Yannis: [00:09:35] It's going to become super easy to do it. And there are many start ups actually working on the VFX side of it. So what we're doing is a bit different in the sense we care about the everyday user, right? So it's an everyday customer. We don't care about... It's very easy to do this in cinema, even with the right budget and the right studio and so on. It's easy to do it. It's harder to do it when you have a phone, when the conditions are not perfect, when you have loads of different kinds of like skin colors, you know, face types, hair. So there are many conditions that you have to consider when you're doing it for a customer, right?

Phillip: [00:10:30] Certainly. Right.

Yannis: [00:10:30] So I think that's a big difference. It's a slightly more difficult task in that sense. But yeah, definitely. I mean and you know, this, I think brings us to the discussion about how terrifying this is, which is a very interesting discussion. Because it's a very fair point. And with all that fake news and this is a reality we're going to have to face and companies like ours will have to deal with it responsibly. And there is I mean, so far I mean, if we want... Do you want to go there? Shall we talk about that?

Phillip: [00:11:11] Yeah. Oh yeah.

Brian: [00:11:11] Yes, that was only one of my questions. Definitely.

Phillip: [00:11:14] So let's actually just ask the question. I mean, is there an ethical boundary at which you think that is it easy to cross? Is it something you've had to establish as a company, as sort of your ethical guidelines? Tell us a little bit about that.

Brian: [00:11:28] And with that sort of what do you see the future of this going? Like what's the ultimate outcome here? And how do you see sort of data privacy playing into that?

Yannis: [00:11:37] So first of all, we're not there yet. Right? So in terms of what we see as the terrifying future, I would say that we're not there yet. And the results that come out that we've all seen online, they're not convincing enough. So if you want to have like Trump or Obama or you know, I'm sure you've seen all those videos, say something they've never said, it's doable now. But it's doable anyway. Right? And with visual effects, with cinema style visual effects. It can be done right? So let's say if North Korea wants to do it, they can do it somehow, right? They can put the money on it and do it. But, yeah, if we're talking about everyday sort of fake creation, we're not quite there yet using this technology. And I haven't seen any convincing from the other companies working on it and research anything that's convincing me that this is going to happen in the next, let's say, a couple of years. Meaning that even the thing we are doing now has very strict conditions. So it's a very global.

Phillip: [00:13:03] Sure. Idealized... Right.

Yannis: [00:13:06] It's not like you cannot use it with any content. You cannot use it with... The scanning we are doing is very specific. So what we are doing, for example, you cannot just start talking convincingly. We cannot just put them in an interview and have him say things that they've never said before. So this requires loads more research, loads more effort and so on. And there is a bigger... Actually this an interesting. I heard that term this week of shallow fakes. So instead of deep fakes, meaning, you know, like simple photoshopping really relabeling of content... Easily created content that's fake, and how bad is actually already right now? Right. So there's an important thing to consider as we're moving forward and the technology becomes faster, simpler and so on. And so, yeah, I mean, we have to be responsible for it. I think for companies like ours, there is something there in the ethical consideration, but there is also an increased need for being responsible for it just for trust reasons. So we want to keep it, keep this technology a closed system, because we want people to trust us with the data. We don't want to open this, so that anyone can use technology for good or for bad. This creates a very awkward precedent. So for us, it's important that this stays safe.

Brian: [00:15:13] Yeah.

Yannis: [00:15:14] Is it going to happen at large? I mean, that's a big question, right? I mean, the other thing that's very important is educating people. I think this is inevitable. I think we will have more fake videos, more fake news, and we see that because we know how easy it is, right? Even as a closed system. So there has to be some education to the public that, you know, things are going to be fake, and the more generated the content is that we see online, the more fake examples we're going to have. So this is going to be a real problem. And there are some ways to sort of use technology to tackle that. Like, for example, let's say down the road, it's not something we're interested in, but let's say we wanted to open the technology up for everyone to use it on any kind of content and so on. There could be possibly a way, and that's something we're looking into actively. How can you use, for example, some sort of watermarking on the technology itself, so that you know that are actually tied any content that comes out that is created is fake and so on. But there are many things to consider though. It's going to be a big thing. Everybody knows it. I mean, it's not like we're saying something that is new. It's already happening. And it's going to it's going to get worse in many ways. So, yeah, we have to be vigilant.

Brian: [00:18:47] What I kind of hear you saying is basically this is not just you. This is and everyone issue really. Like there's an education component that needs to happen in a general sense around deep fakes and technology that sort of merges the realistic, if you will. And if we don't, as a society, start to seriously consider the implications of this, your company's not going to change that. It's something that we need to consider in general.

Yannis: [00:19:24] Exactly.

Brian: [00:19:24] Yeah, that's interesting.

Yannis: [00:19:26] Exactly because I think the problem is not going to come from companies like ours.

Brian: [00:19:30] Right.

Yannis: [00:19:30] I mean, we have zero need for that. We want to keep this technology as much of a closed system as possible. But it can come from other places. So it's very important.

Phillip: [00:19:51] Do you think that that makes you a target in some way as a closed system?Would that make you high value in that you have data that others might want? Or do you think it's easier just to scale their own version of it? And if an attacker would be... I don't really actually know where I'm going with this question. It just sounds like a question I should be asking.

Brian: [00:20:18] Yeah.

Phillip: [00:20:18] It sounds like that makes you a high value target. What do you think about that?

Yannis: [00:20:22] I don't. I don't. Yeah, I don't think that's... I mean, there's nothing to to attack us for right now. So, I mean, the algorithm works great for fashion. So if people want to use it for fashion and they want to attack, it's going to be hard.

Brian: [00:20:42] {laughter}

Phillip: [00:20:42] What are the natural follow on sort of industries that can benefit beyond fashion? Let's sort of talk about the positive or the upside of what a retailer might get out of a product like this. I think a natural one would be sort of health and beauty. Is there a future where you can sort of touch up or modify the appearance of the...?

Yannis: [00:21:05] There is. We're doing some slide experiments on that, but not on the level that you're talking about. It's more about how you create the more photogenic version of the person. You know, thinking about how... Sorry I'm going to something else. But it's an interesting aspect of this thinking about how models provide an aspirational sort of need for the brands and the retailers. So the resulting content has to be a photogenic representation of the user, not just the representation of the user. So especially considering that users will be recording their heads in all sorts of conditions that might not be very photogenic. So there is an aspect of our technology and the things we're developing for that. And yeah, we're not talking about like really enhancing the user in that sense for like beauty and makeup and things like that. I assume that's what you mean, right?

Phillip: [00:22:23] Yeah, I assume that's a natural fit, but there might be others as well.

Yannis: [00:22:27] Yeah. For us the focus with with retail and commerce is see yourself in clothes that you've never worn before.

Brian: [00:22:40] I think that's a big enough problem on its own right now. I mean, definitely your renderings are the best that I've seen. Our listeners should definitely go to SuperPersonal.com and check out the rendering. It's amazing. Actually, in that line of questioning... Actually, one of the things I was curious about is how would a retailer go about implementing your technology right now? Let's say they wanted to use it on their site and have their customers be able to leverage it. Do you have like a services component that's required to implement? How would somebody go about leveraging your technology?

Yannis: [00:23:22] So our approach is a very custom collaboration. So if the brand or retailer wants to work with us, we'll see exactly what they need. Because our vision obviously is we want to create this fully personalized experience, but this has to work for the brand and the retailer in order to scale and so on. So I think initially for us, it would make more sense to have it as a campaign to try it out as a campaign for a specific line and so on, create a very customer experience for the users and go from there and scale from that. So we don't have a one solution fits all sort of approach. We think it's exciting. It's an exciting problem that brands and retailers we think they should engage with and really start exploring and seeing. What is this? What is the use for this right now and five years from now?

Brian: [00:24:33] So this could be good for like a new product launch? Like a very specific set of products or a product line that a particular fashion retailer wants to highlight?

Yannis: [00:24:47] Exactly.

Brian: [00:24:47] Cool. And the technology itself, so are you getting like a 3D render of the clothing itself to create these experiences? Go a little deeper there.

Yannis: [00:25:01] Yeah. So that's a very interesting part of what we do because it's... So the solution we have online and the one that people can see on our website right now is based on actually filming models wearing the clothes. So we consider this to be a styling solution. So it will not be an accurate representation of your body. It's going to be an approximate representation based on how many models we actually shoot wear wearing the clothes. So the illusion is great. And so it's very forgiving. Like all our tests we've been doing it you might even be like 20 kilos off. I don't know what that is in pounds.

Phillip: [00:26:02] It's 40/45 pounds.

Yannis: [00:26:04] Yeah. So even for people that are like 45 pounds off, it's very convincing. But then it's not the fitting solution. So it is more about the style. So we have a new solution because what we're doing now is we're trying to create the right tools for this experience, for our brand or retailer to create. So this new solution we have, which is in partnership with this Hong Kong company called Winswin. Do you know this one?

Phillip: [00:26:40] No.

Brian: [00:26:40] No.

Yannis: [00:26:40] So I'm sure you've seen it. They've build mannequins the change size. So it's one of those companies that build the...

Phillip: [00:26:48] Oh ok.

Yannis: [00:26:49] So we've started an exclusive partnership with them to use their robots for this use. So combining our technology with these robots to change their size, we can actually shoot all the possible sizes. So then this becomes a fitting solution, right? So it's a very specific use, obviously, but this is something we're very excited about because this starts answering some fitting concerns and so on, which is very much needed, we think. And then down the line, I mean, you know, is it going to be 3D rendering of the clothes? This has its advantages and disadvantages, obviously. There are many, many ways to approach this. Right now, we're exploring these two routes. So models and robots. But yeah, we are as a production company, we're very close to the 3D production world. So we're excited to see how this can be combined as well. But we think there's something about actually shooting the garment.

Phillip: [00:28:19] Right.

Yannis: [00:28:19] It's the way it falls, the way... I think I haven't seen many products that are convincing in terms of how the clothes falls on the body, the seams and so on.

Brian: [00:28:39] Right.

Yannis: [00:28:39] It's a very difficult task. And there are companies that are doing amazing work, so I think this is going to happen, but not just today.

Brian: [00:28:52] Not yet. Yeah. So it's very, very smart, actually. I think that the approach you're taking, which is you have a photo realistic render of the face. So why would you pair that with something that's not photo realistic? You're using actual video and images that are shot on real models to sort of connect back to a specific person. It's not actually their bodies. It's not actually a reminder of their body. But in the future, you're going to be using robots from Winswin, which by the way, it was amazing. And I have seen other technologies like this. How real time do you think you could make this at some point? Would you would you even have like a live render of this happen on the spot while you're using technology like the Winswin technology along with the capture of the face? How real-time could you shoot this? Or would not even be necessary? Let's say someone wanted to see their position change or would it just be a static experience or would it be something that they could manipulate down the road?

Yannis: [00:30:21] Yeah. No, for now it's static experience. It's something that has to be planned and produced and then served in essence real time. We don't need some lead time the first time you do it. But then again, it's real time. So, no, this is a very calculated, pre produced piece of content. We don't yeah, we haven't really thought about this to tell you the truth. Of giving sort of this manipulation option. Yeah, it needs some thinking. It's an interesting question. How interactive you want this experience to be like as a next step after you see yourself?

Brian: [00:31:21] Right.

Yannis: [00:31:21] Yeah, it's interesting.

Phillip: [00:31:23] I mean, interesting is I think an understatement. You might be verging on the future. This could be very, very commonplace in the next five to seven years.

Yannis: [00:31:36] That's what we think. Yeah.

Phillip: [00:31:39] Do you think that the consumer expectations right now are being guided by very large players like Amazon? And when they have positive experiences with brands like Amazon or in marketplaces like Amazon, where it's quick and easy and they have some sort of delightful experience, they tend to want to have that experience elsewhere. Is this the type of technology that's really fitting for very large retailers or aspirational retailers and not necessarily for the middle market, direct consumer brand? Like, I guess the question I'm asking is who's the right fit for SuperPersonal at this point?

Yannis: [00:32:25] Right now, it's the smaller players, the smaller size, I would say brands. So not large retailers. In the sense that there's a cost of actually producing the content right now. So if you want scale to like thousands of outfits a week, let's say, this becomes a huge operation, which we're happy to engage with. It's doable. But I don't think this is... We don't know how big retailers would want to engage with this right now. So for now, we think that a brand that really wants to try this out, a brand that believes in personalization and that this is the future like we do, they can create experiences that are actually enabling this. So, yeah, for now, it's the smaller players, I would say, that can benefit from this. Or the bigger players that use it in a smaller capacity. So but yeah, there's no I mean, the cost of it is not excruciating. You know, it's pretty ready for use, considering it's a new technology. It needs some care of how you do it, how you bring it out to the world and so on. It's exciting. We can't wait.

Brian: [00:34:02] That's awesome. I love it. This is something we've been waiting for since really the beginning of the show.

Yannis: [00:34:13] I heard that. Episode 8. Yes, I went back and listened to it.

Phillip: [00:34:19] {laughter} I'm really curious. I'm really curious to hear what happens in the next year and the uptick, because I think there are brands who are looking for differentiators, who probably don't buy into the idea that personalization with air quotes is me being able to show you raincoats when it's raining outside. That is like the absolute worst example that people use in like how I directly market to a customer. This is truly personal. It's super personal. And I love the name, actually.

Brian: [00:34:56] Yeah. Great name.

Yannis: [00:34:56] Thanks, guys. Thank you so much.

Phillip: [00:34:59] If people want to get in touch with you or hear more from you, how can they do that?

Yannis: [00:35:02] So they can send me an email at hello@SuperPersonal.com through the site, or just email me. And we have an app out there, which it's not live. We're going to sort of use it as a testing ground. So it's beta, but people can download it and hopefully we'll be able to serve them soon. Yeah, that's it. If anyone wants to know more, let me know.

Brian: [00:35:33] Awesome. Well, thanks so much for coming on the show. We really appreciate it.

Yannis: [00:35:36] Thank you so much, guys. Love the show. Thanks a lot.

Phillip: [00:35:39] Yeah. Thank you. Appreciate your time. Thanks, Yannis.

Yannis: [00:35:42] Thanks a lot.

‍