Revolutionizing 3D Design at Jordache: DPC, AI & Automation | PI Apparel

Thank you for having me. I want to thank PI for inviting me to speak and my employer for letting me have this presentation. I hope you enjoy this 20 minutes that we're going to be seeing what we have been doing in Georgia in regards to DPC, AI, and automation. I hope that this presentation is informative and educating, um, and entertaining as well. Okay.

Hi, welcome to Jordash PI presentation. The title of this presentation is Deep Dive Journey: The Integration of DPC, AI, and Automation. The objective of this presentation is to outline the journey of the Jordash team over the past 2 years as we've experimented and embraced AI and automation applications to integrate with our DPC processes for the creation of packaging materials and e-commerce images, as well as integrating with the creation of e-learning experiences and support documentation for design development processes to support the upskilling of all our internal team and external partners. And now here with us is Hector Cuevas.

Thank you, Barbara. Okay, so the whole idea was to introduce the presentation with an AI trying to give you a little hint of what is going to come in, right? So if you can see the screen, we have an avatar from Browser, and right next to it is the rendering of the same outfit but with an AI model, right? Basically, we have been trying to test as much as we can to do this type of imagery. I hope that you enjoy the next things that you're going to see. We're going to see a small video of a little bit of nostalgia of Jordache. This video is going to like kind of walk you through of what we have done from the start to where we are now, right? We're going to have another set of videos that are going to be more into the renderings, but let's start with this one first. Okay, so that video was trying to capture all of the different things that Jordans have used for advertisement. For example, you saw from the beginning the video of 1979 till today's date. You saw John Scalia, Elizabeth Hurley, Heidi Klum, and the last person that was the Let's say the embodiment of the brand was Brooke Shields. So we try to keep all of that imagery and that personality in all of the things that we do inside of Georgia Dutch. Georgia Dutch have partnered brands, which is USPA. I actually work very closely with both brands in the use of AI and 3D creations, right? As you can see here, first thing is I'm gonna try to go through all of the different steps and try to define the things that we do for the integration of DPC, AI, and automation, right? So when it comes to creating DPC and AI, every director and every manager should be aware of what are the goals for this type of implementation. Identify the pros and cons. Identify the tasks and deliverables that you're going to be able to generate using the help of AI technologies, right? When you are setting up the goals, you have to think about what is it that you're going to do with it. Are you going to create images for packaging or are you going to create images for e-coms? Or you need to find what are the benefits that you're going to be having once you implement this. You want to reduce the cost of the packaging materials, or you want to actually visualize the garments faster on a human form using AI. Here I have on the slide the— some of the images that we created, right? So, in order to create these images, you can actually see that we actually learn how to create an environment that is a photography environment. We actually learn how to use the 3D software to create some sort of digital photography studio. Actually, you can see here that the tripod doesn't have a camera. So for us, one of the things that we talk in the team is that we have to create the images like we were photographers. So the whole idea is that you have to be able to mimic the whole environment of the photography studio by actually setting up your software to deliver the images like you were actually using the camera. So you can see here we have the settings for aperture, the ISO, and also the aperture depth of view of the camera. For example, all of these settings you have to capture it with the 3D software, set up your field view to be, let's say, 22 degrees to actually have a really clear image. And you can see here that I have the setup from the aerial ground, how the lights are set up. So we have the key light, the fill light, the black light, the backlights. So in the software, you have to be able to create that environment in order to take the picture like you were doing it in the physical space. If you don't do that, you're not going to be able to actually get the best image that you need to, you know, actually create something that is valuable for the brand or valuable for the packaging images that you're going to do or for the e-com platform where you're going to place these images. I believe that if you look at these images, it could be a little bit hard to identify if this is actually real or not. For example, the lady with the vest jacket, that was the first project that we made like around 9, 12 months ago after we went out from the event last year, which just went out to find a way to do this in a way that is actually valuable for us and that we can use it in the regards of the use of the AI images and also in regards on the automation, right? So all of these processes, we are trying to integrate them together to have some benefit out of them and make us run faster in either in the creation of the images or also in the deliverables that we need to use them on, right? So let's go to the next slide. So this is gonna be a continuation of the video where you're gonna see some of the images that we created using AI. All of this is done with browser and other softwares. You're going to see that we actually venture into many different software to create all of this. So let's show this on the video. Okay, thank you. So in order to create these images, we have to learn how to do digital puppetry. We have to find a way to create as much poses as we can, and the poses have to be trying to mimic the real models that we usually do in the traditional methods, right? So it takes quite a lot of time to understand that if you have to mimic a person, that means that you have to go even from, you know, the fingers to the hair, all of the positions. You have to kind of find a way to mimic this. And I think this portrayed very well the how we can actually look at the avatar next to the actual AI rendering. It could be very, let's say, extreme, the detail that you have to get. But once you find a way to associate the body of the avatar to an AI model, they find finding the right model that they actually looks okay with the clothing that you are creating for the brand. It has to identify itself with the brand. It cannot be something that it looks, let's say, unnatural. And that is actually something that takes a lot of time to realize, that if you create an image and it doesn't look natural, people are going to reject it. People are not going to look at it. They're going to just feel that this is something not appealing to them, right? So selecting the AI model Selecting the right body, selecting everything that needs to be transparent with the brand is essential to do this, right? So here I'm having some images from the digital asset management that we do in Jordache. First thing, you should maintain a reporting of all of the assets that you have. You need to be able to count them. You need to know how many do you have and what they are doing for your brand. Like, if you make 300 garments in 3D and they just buy 20, maybe that's bad luck. But at the same time, you also were able to do all of the bad samples that you needed to do beforehand and get the benefit of having the asset to be able to help you to get to that point. To get to the best garment that you can make and get to the best position that you can get. All right? So here you have on the left the reporting tools. You need to be able to measure how many were bought, how many were not bought, how many rounds did you make for them. For example, if you see the graphic, it has all of these little numbers, and that's counting how many tries did you do for each 3D. In the center, we have the AI model faces. So that becomes also another thing that is part of your inventory now. Now you need to have all of the different faces that you're going to use, the ones that belong to the brand that you're going to use, associate which size of the mannequin matches that face, what poses will look good for that model. The 3D materials is also an essential part. You need to have a library with all of the 3D materials, with all of the studs, buttons, buckles, shoes, accessories, everything that you need to have to make a great image. It will depend on how good is your library. If your library is not good, it's not going to be good. Is going to be something that is not going to be efficient to what you want to do, right? The blocks. You need to have a good library of blocks. If you don't have a good library of blocks, it's also going to damage the process of making this AI integration. Okay? So let's move on to the next slide. Maintain a list of prompts. This is something that it took me a while to kind of digest that we will be able to create images in AI with a prompt. But as you can see here, this is some of the examples that is available to everybody. Everybody knows what brand I'm using here to create these images. You can see that if you type a prompt, you can change the face of a model. Without too much hassle. It takes a few seconds. You can actually generate so many faces in little time in case that you want to actually switch the face of the avatar. In this case, we are having this kid, right? And the idea is to be able to render and then switch the face to a kid in order to match the whole objective of the image, right? This is a kid, Carmen, and we have been having a lot of trouble to find the way to generate images for kids using AI. So, so far, this is a quick solution to try to find a way to resemble the kid using generative generative AI technologies using Adobe Suite, right? So let's move on. Okay, the changes on the process. You have to identify what are the processes that need change in order to implement AI in automation. You have to stay true to what your brand is doing because if you start making all of these changes in going so forward, you're going to disconnect to where you actually are. You have to stay true to the brand itself, to the identity of the brand, and the purpose of implementing this cannot derail where the brand is right now. Okay? Maybe you want to create an integration with AI to deliver your assets faster. Or maybe you want to create the assets and also be able to deliver them to the individuals that are going to be working with them in a faster way. In order to do those things, you have to be able to learn how to use your PLM as a remote control. It will be very impossible to do all of this without a PLM system. The PLM system, you can actually integrate it with APIs and webhooks to be able to do automation with your PLM. If you are able to do automation with your PLM, you need to kind of prepare yourself, prepare yourself to change your processes, change your equipment, train the teams. The teams have to accept that this is something that is going to help them. Open your mind that this is not something that is going to replace you. It's just that you have to find a way to insert it into your process. For me, AI and automation is like a second head or a third arm. Some people feel a little bit bad about this because probably you think, oh, Maybe you want to change everything just using AI. Maybe you want to take my job, or maybe you just want to take everything, all control of the whole process just by using this. No, if you implement AI in automations, you're going to be able to free yourself from some of the horrors, some of the things that takes a lot of time for you to finish a task or a project that you are making. Okay? You have to also be able to evaluate your equipment because when you do this type of imagery or this type of project, a regular computer won't work. You need to have the best that you can afford. For example, the team has been able to achieve great things using AI since we have the right equipment to do the correct process for DPC, but also to be able to render faster and better. All our equipment is at least $7,000. You're not going to make it with a small laptop. You have to be able to render 300 images in half an hour and be able to move on and do 300 more till you get what you need. Okay. If you want to scan the code for later, if you want to ask questions, you can do it now. Okay. So automating your workflows. This is a part of my presentation that is not so friendly, not so beautiful. Maybe some people will not appreciate this, but for me, automating my workflow is essential for me to deliver what I need to deliver to my peers or to my boss or to somebody that is outside from the company in a faster way. For example, here you see task send images to selected members without opening my— an email application trigger plus custom action equals task done. After the user clicks on the save button, automatically the images get delivered to specific team members. If you think about it, when you send an email and you have to gather the images, you have to type the email, you have to introduce the idea, okay, these images are this, you have to put the subject and everything. If you have to do that for 10 different styles, you can spend 2 hours something that can be done in minutes or in seconds. If you see my diagram here, I have 3 sections. One that says trigger, another one that says action, and the last one says task done. So in the first one you see user, PLM, the arrow moving forward, click on the save button. So I'm kind of glorifying the save button. Why? Because once we integrate the webhooks or the API with the PLM, everything becomes very powerful. Every action that you do with your PLM could become a trigger for something extra, right? So you can see underneath the save button, it has the purple rectangle webhook. It's listening to the PLM system. And right underneath we have automation platform, right? So there's automation platforms that you can integrate with your PLM system. And once you integrate your PLM system with the automation platform, there's things that you don't even have to think about that they are already done. For example, today I saw a couple of emails from a peer that is in China and The same method to deliver the images was done the same way that you see here. She uploaded the images to the PLM system. She clicked on Save. Everybody got the images. No email, oh, these are the images that you have to receive to do this and this and that. No. You can actually make this type of automation to a lot of different processes, right? Just by connecting your automation platform with your PLM, with a webhook. The email gets sent automatically, and then whoever has to receive it is going to be happy instead of, oh, where are the images? Can you send me the images? Done. I'm trying my best to make the life easier for my peers that are in China and make my peers in New York as well feel comfortable that they're going to get what they need faster. Without hassle, right? Okay. So, this is a little bit more technical, but it's gratifying once you understand it. So, we have to customize and start automizing your workflow. Is the best that I was able to study for this whole year was to try to integrate the automation platform with the PLM, right? But in order to do that, you need to understand the basics of the automation. The webhook is mostly a URL that you can get from your automation platform, and that URL you can place inside your PLM. Once you place it inside your PLM APIs, you're going to be able to do the same automation that I just showed you before. Um, you have an option also to do the automation with APIs itself, but that will become much more difficult. I don't even understand how to do it. So the only way that I was able to, to actually get my hands on automation was by understanding the webhooks. And the webhooks is the URL that I mentioned that you can actually place inside your own PLM. It's listening to every action that your team member or do you are doing. Let's say that I add a comment into a style, and if I have the right webhook in the right place on the website where the comment is added, I can actually trigger something else out of it. So the webhook is— it gives you the capability to do that. Receive webhook, as you can see in the diagram. Valid signature, process data, send response. At the bottom, you can see reject request. So you can set up that once you have your webhook installed in your own PLM, create conditions for you to be able to receive the deliverables that you want to receive, create a spreadsheet without you opening Google Sheets or Excel, send an email with the images, send a notification. So many things that you can do by using this. If you use APIs, protocols to do this, it's going to come to a point that you need to have a developer. I've been bothering the guys from Beproduct, which is the PLM that we use. Quite a lot to the point that I had like probably like 4 months of sessions 2 times a week just to get this going because after the race from the last PI last year, I was— I talked to my boss, we have to find a way to implement this because it's going to be beneficial for everybody. And as we go along, We have seen good results. I'm satisfied myself because I use it for things that are very clerical. For example, I do a lot of onboarding. Every time that somebody comes in the company and in the design area, I have to train them how to use the PLM, how to work with the digital tools that we have. To do 3D or to do something else. So that way, by doing this and using webhooks, I'm able to get away from a lot of things that I can just set up and once somebody comes in, I can actually send them the onboarding documentation without me grabbing everything, which is an email that actually can take 20 minutes to prepare and so many other things just by using these methods, right? Seems like I'm running out of time, but I'm going to try to explain to you. You need to find a way to use the automation tools to be able to connect this, right? You need to understand what platforms are best for you. Probably a platform that has less coding is best for you, or probably a platform that has more coding is better for you, I would recommend to use something that has AI, for example, like Zapier or Make. Those two platforms are very user-friendly, probably Make more than Zapier. And you have to identify what are the things that you're going to need to change to make this adoption of technologies. You have to Pick the ones that work best for you, the one that actually makes everything flow easily, right? So I'm going to move to the next slide before I run out of time. And you can see here that I have a diagram on how to do the basic steps. So you have to identify what is the trigger for the action that you want to automatize, right? Define the trigger, define the custom action, meaning what action is going to trigger the next step, test the result, repeat the testing, report, measure what was the result, check the logs if there's a problem, and start all over again. So you have to do that many times in order to make this work. Automation test cycle in the next graphic, you can see here that I'm trying to explain where are the steps to do the basic testing, okay? So trigger, custom action, test, and the delivery results. If it's success, you can just retrigger again. If it fails, you actually have to go and test the trigger, test the custom action, test your conditions, and then move on to the cycle again. Check if anything is contradicting the workflow. Fix the issues and come back into the retrigger. On the one on the— all the way to the right, the webhook is listening, right? So you already went through the first steps. Everything is set up. So once you are there, you just have to make sure that the flow goes perfect or as much as you can, no troubles if, if the information gets to the right person, if the email grabs the right images to send to the next person, if the actual Excel form that you're trying to make without you open Excel is actually populating all of the information that you need to populate. Okay, so let's move to the last one. Okay, so in here, very simple, probably for some people this this is not interesting, but for me, it makes the day. This is the example. Somebody's actually putting the images in the PLM system, clicking on save, and then after probably a minute or 30 seconds, you're going to get it on your email. You just have to make sure that you refresh and you make sure that everything is actually properly going in your own webhook. And you should be able to have everything that you need. So you can see here these are the same images that I had before, and right now reloading the page and I'm going to get my email. So if you think about this, if you have to send 50 images, This will make your day because instead of you doing it one by one, it's done automatically. Okay? If you have any questions, I'm free to answer them. Just let me know. Yes, sir. Austin, are you rendering browser or you're rendering outside of browser to get the 4:3? Oh no, everything is in browser. Yeah, we try multiple softwares, like I said, probably 20 or more, not only for DPC, probably some compositing 3D imaging software for architectures and so on, just to test how the light works outside from Braswer. But thank God Braswer got to the point that we started to trust it more after they installed the lighting and so on. So the digital studio, we started by creating GLBs and put it into another software, creating the digital studio, taking the picture, Try many times, but at the same time, some things that you do outside from Browser, you cannot really mimic inside from Browser. But I found a way. If you look at the picture, the first picture, we have the AI model sitting on the share. So I was able to bring the share inside Browser, which is something that you probably will need another software to do and see the model there. It took a couple of tries, but it was successful. [Speaker] Cloud rendering? [Speaker] Both. I found that if you have the right computer, you kind of don't need to do cloud rendering because if your computer has the right graphic card and the right RAM, For example, the one that I use has 128 RAM. 64 won't do it for me. 128 has done the job. I can spit out 30 images probably in 5 minutes, which is maybe more if you do the cloud rendering and the activity of going up and down, you know. It all depends what is more comfortable for you. But I found that using your computer, if you have the right computer, Maybe you don't need that. I'm on a crash. I want— Okay. Thank you, guys.