Latest / Talk Commerce / AI and AR Drive Beauty Tech Innovation at Perfect Corp With Wayne Liu at Shop Talk
Transcript
- Speaker 2 (00:00.974) All right. All right. So it's conversational. I'll do an introduction. you're with Perfect. You're with the Perfect Corp. Perfect Corp. All right. And it's Wayne, Louis. Lou. OK. the perfect corp. Wayne. do. Yeah, can do. So I'll introduce you and then I'll let you do introduction for yourself. And then we'll go into content. We'll try 10 to 15 minutes of content. And at the end, I give everybody a chance to do a shameless plug so you can promote anything you want. And is there anything you'd like me to today? Well, I don't know how much you know our company, but basically we do AI and AR, and help this retailer and brand focus on. So that's why I guess if you want to focus a little bit, we can talk about the AI, how AI help this retailer, and then some of the results we see. Okay. And AR, for AR, are you referring to accounts receivable or augmented reality? Yeah, that's what I thought. Just making sure. Good. All right. So we'll get started. He's already hit record. So all right. So here we go. Speaker 1 (01:10.702) Yeah. Speaker 2 (01:19.852) All right, welcome to this special episode of Talk Commerce live from Shop Talk. Today I have Wayne Lu. Wayne is the co-founder and chief growth officer at Perfect Corp. The Perfect Corp. Yeah, Wayne, go ahead, do an introduction for yourself. Tell us your day-to-day role and one of your passions. Absolutely, thank you, thank you for having me, Brent. I'm from Perfect Corp, so the Perfect Corp is a technology company, been in business for about 10 years. I co-founded 10 years ago, so this year actually is our 10th anniversary. So basically our technology is AR and AI, so our service industry primarily is beauty, and also getting to fashion. So our technology, we're using AR to help people to virtually try some of the products which they have a difficult to try at home. For example, makeup, all these lipsticks, eyelashes, and then we get into skincare. So we are using AI to analyze your skin. It's a very, very intensive deep learning, machine learning. We have about like 90,000 data, so we can test your skin in five seconds, and then you know 14 different concerns, 14 different concerns of your skin, and then the brain can do product recommendation. Yeah, and then all get into the fashion. and do all these trials. So as I say, we've been working with more than 700 plus brand and retailer, including all these big name Walmart, Sephora, then, know, S &L, the LVMH. So that's pretty much where we are. And does your technology, are you doing in-person retail or and e-commerce? Speaker 1 (03:00.686) Yeah, so basically we're a software company. So if you license our software, we can license this form of API or SDK. So about 70 % of our customers actually integrate our technology into their e-commerce site. But some of them also can put into their kiosk. So they take SDK, put in this kiosk. that's why some of the retailers have that in store. explain how I'm intrigued with the skin care analogy and how that would work online. Is it through photos and things like that or is that a solution that has to be in store? Yeah, so that's amazing about this technology. Basically, we start with a photo, but now we can do live. So the technology is basically it's machine learning. So we work in the past six years. work with all this. We start with some brand, like Neutrogena, Lourish, Perse. So they have lots of data for us. So we start training our engine. So that's why, actually, again, it's the same. It's API or it's SDK. So we can use using just the camera from a PC. So you open up a website and then the camera is on and then scan your faces and then just in five seconds tell you how your skin look like. And we also have a so-called more like an off-the-shelf product which is more like in the form of an app. So the aesthetician, dermatologist, they can download into the app and then their app turn into a skin analyzer. So that's the software. we don't really, it's not really It depends on any hardware. Yeah, hardware diagnose. Speaker 2 (04:45.838) I know that people think, people have the impression that ChatGPT just came out, which it did, but machine learning has been around for years, right? You started 10 years ago. Did you start with another system like IBM, Watson, or something like that to help you to get you into that space, or have you developed a solution from the ground up? Yeah, so that's a great question. So basically we understand that AI, have analytic AI and the generative AI. So the ChetGBD is more on the generative AI side. So we start, because all this machine learning is about analytic AI, so we have lots of the data. So we have lots of data, we start training the machine. This is all pretty, we call it classic AI. So you're training with a neural network, we use a lot of convolutional neural network. So at that time, we doesn't really need to have any so-called transformer, this type of thing. So yeah, so that's how we develop the skincare. So analyze the data and then can do prediction. However, now we are in the era of generative AI. So we can do simulation. In that part, AI start to create things. So to create what if, right? If you do this, what the skin will look like. So that one, we have a generative. we do have a product called Perfect GPT. license the open AI API and create our own rack. And then that's how we start with perfect GPT, specifically the GPT for beauty level. Yeah, you know, I have another friend who started an AI company or machine learning company about the same time as you and they do talk about the neuro network. Can you explain that to our listeners what that means? Yeah. Speaker 1 (06:39.662) So basically, how machines learn. So that's a fundamental question. First is a machine doesn't really learn like a human because they don't have eyes. So that's why we need to start with computer vision. when we send a photo to a machine, and then we need to train them, what's the edges, so what's the color. It's a layer. So basically, we are using many try and error. So we show computer or show this AI engine a photo. For example, it's a cat. And we expect them to learn. It's a cat. If it's right, we say right. If it's not right, go back to learn. So these kind of iterations, try and narrow. And therefore, neural network is because just like a simulation of a human's brain, there are many, many different nodes. So you have a brand cell and brand cell connect with all this of the neural network. So that's how they connect it. So you try to develop an algorithm similar to human brain. start doing the convolution. Convolution means a layer. Because, for example, in our case, it's color. Color is a many layer. So you cannot just do it once. You have to do it repeatedly. So there's convolute. So repeatedly, you start getting machine to learn. So that's how. to put it into a very high level explanation. And so you had mentioned some fashion brands. Is that the best fit for your solution, or is really anybody that needs some help in terms of either in-store or online would use Perfect? Speaker 1 (08:25.42) Ideally, in a nutshell, that's the case. But the thing is that we focus on beauty and fashion. That's the reason. For the beauty, because we have probably, we'll say, one of the largest color library. We have about... 800,000 different skill, we call it skill. It's all color, color library from different brand and retailer. they, you know, the color, the precise of the color, it makes this virtual triumph real. Okay, so that's the one. Ideally, you can think probably we can have furniture, right? So it's also works. But then create a 3D model can be a little bit challenging. Because our specialty is really create a 3D model for this accessory, right, eyewear. possible. But just like as a business, we just focus on this industry. So you started off with AR and of course I got it wrong right away. I said something different than augmented reality. What do you see that's happening that's exciting, that's new in the space compared to what you saw five years ago? Yeah, I think AR is still very exciting space, but only for certain industry. It's not for everyone. What we see here is more like a quadrangle. So if your product can be standardized, for example, color, for example, furniture, it can be standardized. And then that's easier to create a model. But if you are selling an onion, which is a little bit difficult to standardize it, so they're probably not Speaker 1 (09:58.128) AR is not for you. And then the second requirement is you have to have a variation. For example, makeup is a different color. For the furniture, of course, they have a different color, a different shape. So that's why they will make a virtual triumph fun. So that's why if your industry has a, product can be standardized and then has many variation. I guess that's a good place for the AR. So that's why, then also AR, just to answer your question, we see tremendous returns. It's not just something for fun, it's for business issues as a tool. So if our customers are using virtual triumph, right, so they get a good conversion. And also they reduce the product return because they get the right color. And the consequence of that is it makes your business more sustainable. I think one of the things that set e-commerce apart maybe 10 years ago was just mobile commerce where people could use it on their phone. Do you see augmented reality as that next breakthrough for just regular retailers that are using it? Yeah, so I think right now retailer is already using AR with the phone. it's because the phone has a very powerful core, can do the computation and also the camera. Another thing here is that we get very excited about this AR glasses. Because AR glasses, this company tried many times. I guess right now we have a chance to make it right. And then the good thing about AR, the AR glasses is because you are using the first viewer's angle. But you phone as like a third viewer. So that's why I guess if we can get this AR glasses right, it will dramatically help the retailer. Especially on the in-store side. Speaker 2 (11:56.248) When you say, you mean wearing a virtual reality headset and then people can visualize what things are? OK. Would that be like the Apple product that they sell for an exorbitant amount of money? That's why I said Apple's one probably just for fun for this You know like a tech savvy people, but what I'm referring to is really the meta the meta and Ravens one Yeah, so just like a about $300 something and then they get whatever people need it Yeah, you don't have you don't need to have a very very powerful function. Just something is a good enough Do you think in terms of how people in e-commerce are using this technology that it's going to become available to more and more people as it gets more widely spread? And a lot of these, like I think you said, that a lot of the bigger companies are already using augmented reality. But we don't see the mom and pop shop, like the corner little store that's selling furniture. Maybe they can't afford to get into it. Is it going to become more democratic in terms of how they the access to that technology. Yeah, I technology wise is definitely possible. But what I see here is again the 3D object. So if you have an easy way to create a 3D object, I think they will get this one more popular. And then the key thing for this 3D object is right now you are using AI to help. So you probably get just a couple of three photos of your product you are trying to sell. And then the AI can Speaker 1 (13:36.032) rendering this photo and then create a 3D model. So that's not the most accurate one, but it's possible. So if we do that, I guess the small SMB company can definitely do it. So right now we have a pretty success on the eyewear. So we don't need to have a real 3D rendering. We can have three photos and then using AI to render this for them. So if you are missing part, AI will compensate. And would just fit clothing fit work for people? Maybe they would just have their iPad that's looking at them and then they could theoretically fit clothing on them. And would it then work to send or to add to buy that piece of clothing in the size that the AR model would see them in? Yeah, so for the clothing, the most challenging part, I think the technology is there, but most of the challenging part is to do your body measurement. Because if you do, say, like makeup or skin, you don't have anything to cover your faces. But for your clothes, it's already been your body being closed. So we don't know what your body look like. So that's why if you want to do a very precise fit, they take some process. But if you want to do something just to take a look at this style, the color, if you feel it. I think the technology definitely is already there. So we're at the Spring Shop Talk 2025. What are you seeing that you're excited about, either in your space or in another space that you see that's innovative today? Speaker 1 (15:10.69) Yeah, I think, again, it's a cliche, right? So just like, you know, it's a CES, the AI. The exciting thing is about AI. The frustrating thing is about AI. Because everybody provides AI. But I will say probably one. out of 10 is a good AI. Because what I define a good AI is really AI try to solve the problem. It's not clear, not a layer of problem. So I guess people talk about AI, but you need to figure out first if this AI really solving my problem and how I'm going to implement it. Some of the AI solution which I saw is good, but it's very difficult to implement. as you say, probably like the global, this is a multi-country national company you can implement it but small companies they have no way to assess it so that's why if you create a good AI, solve problems and affordable I think there will be a very exciting thing I saw some of this and some of it potentially can be in that area so that's why You mentioned earlier about the trained model, would be machine learning, then a generative AI, which is what we see as chat GPT. And I know that you can make your own GPT, but it doesn't necessarily learn day to day. Do you make a distinguish between machine learning and generative AI still in how you reference and build those models? Yeah, so for the machine learning, machine learning actually is the core, it's the foundation. So even this GBT, they do machine learning, but the way they do machine learning is they do so-called unsupervised learning. They just go ahead and learn by themselves. They go to internet to grab all this material and learn. For the machine, we call the classic machine learning is we set the boundary. We train them, and then we want them to learn what Speaker 1 (17:08.695) So the good thing, the exciting thing about right now is we combine these two together. For example, our perfect GPT, we license open eyes, larger language model because it doesn't make sense to respond well, so that's why we license them. But for some very specific beauty related, we have our own trend engine. So if the customer asks a specific question, so the machine will only get into the answer from that specific a trend database. Okay, so that's how we architect this so-called private GPT, if you like. The DeepSeq came out, or it's been out, but in January it got a lot of press around some vulnerabilities and maybe some copying of other models. But the exciting part about DeepSeq is the way that it distributed the model over many nodes. Do you see that as something that all the other machine learning platforms should be looking at to reduce cost? Yes, so I will say it really depends. So if you have a, if you like open AI or like you know probably like Amazon or all these companies, you have a resource right so you probably should train a large language model. Okay and then the DeepStick actually is based on large language model. They do the distillation right so they take a part of that. the model and then they just trend it called the X-Persistent. So that's a small language model. But the thing is, the thing about this, you have to have a company to start with a larger language model first. Those company can start using this distill, like a discreet, right, as you mentioned, this different note. So that's why I guess, still some company need to do the hard work. And then probably the rest can take, can leverage that. So that's probably will be the model moving forward. Speaker 2 (19:03.278) Is there anything that Perfect has coming out that you can tell us about that is exciting that's coming up? Yeah, so a couple of things. First one is, as I said, it's the perfect GBT. We have this technology last year, but we will officially launch very soon this year. So first is we'll put it into our app. We have our own app, which is very active. And then we can start, you know, do this conversation, and then more like expert system to talk to it. And then also on the skincare side, we have some very exciting development. First is we do HD version, high definition. So currently we do a standard definition. But right now if you do the HD you can see more. So you see the wrinkle, not only just a wrinkle, we see like five different type of wrinkle because that's how you trend with a high definition camera. And then also we're coming out with something called aesthetic simulator, which can solve a problem for this plastic surgeon. Because usually what happen is the patient, don't know what happened after they do this surgery. So that's why we are using AI, of course AR, try to mimic, try to simulate by their own faces. And then so if you do this surgery, probably like you know, put some filler, Botox, and what will happen. Okay, so that's why it's a simulation. Instead of looking at the model's faces, look at your own faces. Well, that's amazing. Yeah. That's great. So Wayne, we have a few more minutes. As I close out the podcast, I give everybody a chance to do a shameless plug about anything they'd like. What would you like to plug or promote today? Speaker 1 (20:43.02) Yeah, so I'll say, know, Perfect Core, we are in the very exciting business. We are doing the technology and the fashion of beauty. We are in the intersection of technology and the beauty. So it's very exciting. We have a lot of products. You can go to www.perfectcore.com and try the demos, though. Whatever I say, you can just try it there. And then, yeah, so and then I... Just keep up with us. You can go to Linking and make a connection with me or Perfect Calls. We release lots of new exciting products. That's great. Wayne Liu is the co-founder of Perfect Corp. Thank you so much for being here today.