How Raspberry AI is Transforming Fashion Design | PI Apparel

Uh, I'm Cheryl. I'm the founder and CEO of Raspberry AI, and we're a generative AI platform focused, um, on the fashion industry. So apparel, footwear, and accessories. And we primarily serve the design teams, um, help them go from concept ideation to tech packs, all in one AI platform. And we recently expanded to some marketing use cases as well for internal presentations or external-facing marketing assets. And we're so privileged to have one of our wonderful clients, Kunal Shah, who's an SVP of product development and technology at Kelwood Company. So together we'll be talking about how Kelwood is using Raspberry in their workflows and how other clients have been using Raspberry to save on sample cost, increase speed to market, and really superpower their design process. Hi. All right. So we're going to cover 3 main agenda items today. The first one is how to use generative AI in your current workflows. The second is how companies we've worked with have defined success in their AI journey. And the third is an overview of all the— our key generative AI features that we offer today. So this is the whole gamut of what Raspberry AI can offer. As you can see, we're really serving the entire end-to-end creative process. So starting with that initial consumer and market research, right? Understanding what the trends are, understanding what consumers want to buy through Google data, through TikTok data, through Instagram data, and then going through the typical design process of ideating, so generating potential collections, generating prints, taking your reference images and generating variations of those, and taking your sketches and rendering them into product images, editing those to the final prototype, getting some initial technical design assets, so 2D flat sketches that are Illustrator files. And then in the yellow section, these are new features to Raspberry, which a lot of our clients have been taking advantage of, which is going beyond the product, the garment itself, creating lookbooks, creating really compelling on-body with scenery. Imagine, you know, your Saint-Tropez photo shoot, and then also e-commerce assets. So multi-view showing the front, back, sides of all your garments, all generated with AI. And here's some of the clients that we are working with today. As I mentioned, we are 100% focused on the fashion industry. So all of our clients are in this space. And we serve womenswear, menswear, mass market, premium, luxury, kids wear, baby wear. So really across all different segments, Raspberry can be very helpful. In your design process. And I'll bring Kunal to the stage now to talk about how Kelwood has been leveraging Raspberry over the past year.

Thanks, Cheryl. Hi, everyone. Um, so I want to start off by saying when you delve into the world of AI, a product means different things for different companies. So Cheryl and her team were kind enough to beta test the software with us for 3 months last year. And then we brought in, based on that cross-section of users— so number 1 is speed to market. For the brands that need to really get their product in front of retailers sooner rather than later, using Raspberry, using the feature like Sketch to Render that generates photoreal assets so we could do presentations in a day or two versus a few weeks has been, you know, really helpful in speeding up our turn. One of the main areas we find it useful in is they have a print generator that has been really, really good as far as the usability. We even have started using it for graphics, and there's two ways you can use it. You either input a reference image or you can do the prompt search method, and Either way, we found it really helpful in generating prints in seconds sometimes. And within like 5 minutes, you could have a usable asset in separation and repeat, which is pretty amazing. Also, for the brands that are owned at Kelwood, because we do some private label as well, the internal process can be sped along with the speed at which Raspberry is able to generate assets. So that, that's very, very helpful. I mean, using the tool in selective areas, we've seen an improvement of, you know, 2 to 4 times efficiency in terms of our speed. And that's where I want to highlight the real focus of a software like this should be to make your teams efficient and agile. It's really about, you know, becoming more, more efficient with the team you have. I am not a big proponent of AI, you know, replacing humans, but if you're trying to scale your business and maintain your headcount, this is a great way to do it. And lastly but not the least, we're exploring the use of Raspberry's marketing tools, as Cheryl Sheryl mentioned, for e-commerce and product shots, especially using these assets where you need to do a photo shoot in a location-specific area. We're, you know, launching a new brand at the company where we hope to really incorporate this tool and use it to a larger degree. Thank you.

Thank you so much, Kunal. So we're going to cover a couple of other case studies from different types of clients that we work with. So Kelwood is an amazing brand. They have their own brands, and they also sell to a lot of great retailers. One of the case studies I wanted to highlight is from one of our UK-based retailers. They sell apparel, but a lot of other things. And they're at $9 billion in revenue. What they were really looking for when they were searching for an AI design solution was to solve a few pain points. Number one, trend research and inspiration research was very time-consuming for them and took them multiple weeks every single season. Sketch iterations and the sketch review process and line review process were also extremely time-consuming and extremely expensive for them. Before they started using Raspberry, they were actually sampling on average 3 to 4 rounds per, per SKU. And with using Raspberry, they were able to increase their first-time-right samples by 50% and on average reduce their sampling rounds from 3 to 4 to 1. So that really helped them cut sample costs down by 60%, which was— which far exceeded both their and our expectations, to be honest. And I think just really shows the power of an AI tool like Raspberry where if you're cutting your sample rounds from 3 to 1, you're also cutting down 3 to 4 weeks of sampling throughout each round, right? The other things that they wanted to improve with Raspberry was just decreasing the manual work. A lot of, you know, copy and pasting into Miro and copy and pasting into different presentations and presenting in all sorts of different ways to different stakeholders. And all of that they were able to do with Raspberry in less than 3 months. Since starting with them in September, we've now are going full-fledged into their, their entire designer base. So you can have a very rapid AI adoption journey. The next slide is about the manufacturing client. I know a number of you guys here may be manufacturers or work very, very closely with your manufacturers. So this manufacturer is based in India. They're one of the largest manufacturers in the world. And they serve clients like Walmart, Gap, Knitwell, Kohl's, Amazon, American Eagle. For them, I think very similarly to Kunal and Kelwood, they have a massive prints business. And they partner with some Italian print studios. But it's very time-consuming to create seamless prints repeatable and separated by hand. And what they were able to do with Raspberry is they're not losing any of the artisanship or artistry of their prints, but they're able to start to generate lots of different options, like 10 times as many options as their team could do manually by hand. And then really focus the manual time on the prints that were most desired by their clients, and therefore also massively improve the speed of product iteration. For them, they were able to cut the new print development down from an average of 3 days to 1 hour per print. And they were also able to reduce the product development requirement in both time and cost by 40% to 50%. And I think most exciting for this manufacturing client is— and also brands who are selling wholesale, right— is the buyer uptake. When you're presenting to Macy's or Bloomingdale's or Dick's or whatever, what have you, for them, they were actually able to increase their buyer uptake from 30% to 50%, which is really massive and really impactful for their business and for sure covers the ROI of any software cost. All right. So those were the case studies. I know a lot of folks here are probably very visual, so we're going to go through a lot— the different features in Raspberry in a very short amount of time. Okay. So— there we go. What we want to show in this graphic is just the end-to-end workflow of Raspberry. This is one one of many, many workflows you could have, but I think a very common one for our users, which is we started with a sketch. We rendered it in that red ombre. And then we put it on body on one of the top tennis players in the world, Naomi Osaka. And then we put her in Roland Garros. And all of this is done at our computer. All these are generated in 15 seconds or less. And beyond apparel, like we— I— like I said earlier, we do serve the footwear and accessories categories as well. So same thing here, you know, you start with your sketch and then instantly you can get a photoreal product image. And then you can put that photoreal product image on a model. So we're replacing these brown boots with these black patent leather boots. And then we're putting her in a scene. So here we're doing a desert Dirt Road. And it just really elevates this piece, right, from that initial sketch, that black and white, really bare bones skinny sketch to this asset, I think just really speaks volumes to the vision that the designer had in mind. And now we're going to get into some of the features here. So Sketch to Render, and this is, I think, One of the key use cases in AI design is sketch to render. So this is where you can really bring your concepts to life. The left-hand side is the input. It's your sketch. It can be hand-drawn like these. You would pair it with a prompt. So for example, like navy leather collared men's jacket or dark green hobo style handbag. And it really turns it photoreal. And this is not only like a beautiful asset that communicates your vision, it really has practical business implications and matters to your bottom line because what this sketch now replaces or this render replaces is that traditional line review process that you guys had with sketches. And it's what is really able to help save up to 60% on sample cost and sample time and reduce the number of iterations. And I think through this example, what's really interesting here for concepting is You're able to take that one sketch, right, but concept it through a black patent leather, and then concept it through a sparkly sequin, and then concept it through a cream shearling, and then also refine the cream shearling down to the shoelaces and the color of the, the sole. So you're really able to explore a lot of combinations that we wouldn't be able to sample physically just because of, you know, time and cost constraints. And I don't know, Kunal, if you want to say a few words about how you guys are leveraging Sketch to Render.

So just to highlight further, Sketch to Render is a great tool in terms of improving speed to market. It's a great way of using your Illustrator, Photoshop, CAD, and bringing it to life, making it more of a 3D-like asset. It is not a 3D asset, but it's 3D-like. Where the fabrics also are, you know, they reflect what they're supposed to be. And if you're trying to do a bias-cut charmeuse dress, there is a way to, you know, really bring that to life. So it's a way to really look at what you're presenting and be able to present an entire line that could take weeks in a couple of days to the retailer. So I think in terms of just helping design really see their vision come to life and helping also to present it to the retailer, it's a really, really great way to do that. And going from a sketch, a black and white sketch, generating your own print, plugging it in, you know, putting it through sketch to render as Cheryl said, and then putting it on an AI-generated model. I think it's a fantastic presentation tool, and some of our areas that are using it have really benefited from that. Thank you.

One more thing I'll say about Sketch to Render, because it's one of my favorite tools, is I think it— what we've heard from clients is that it's really elevated a lot of the internal line review discussions, right? There's no more confusion about what this should look like or what material this is. You can render exactly. You can put in a fabric swatch in Raspberry. So it's all clear to everyone what that final result is. And it's truly changed the line review conversations. So beyond starting from sketches, you can also start in a number of other ways. So this highlights starting from a prompt itself. So just a couple words, really whatever you're thinking for the line. So here we're thinking about a lilac, satin top with a large bow detail at the bust. So we're typing that in, and Raspberry will generate all sorts of different options for this kind of garment. And the great thing about AI is it's not going to give you just 3 options. You can click Generate, Regenerate a million times, and you'll get 3, 4 million different options. And they're not searching from Google. It's really a new— product generated each time. And we also have software within Raspbery to help you measure how similar the outputs are to other things on Google Shopping so that there's really much, much lower risk of any infringement issue. The other way you can start generating more prototypes is with a reference image. You can type in certain changes you want to make, so And we're starting with this Versace boot and then we're saying we want to add a chrome accent. And it's, it's adding that automatically. We don't have to go into Illustrator. We don't have to go into Photoshop or changing it to a metallic finish leather. And it's just automatically doing all that work for you to get to a great prototype. And here's some examples for apparel. So again, you put in a reference image. Maybe this is a bestseller from your current line, maybe it's something you're inspired by. And then you can really just type in what you want to change about it. So here I'm asking to change it to satin. And I'm asking to add a large teal ruffle detail. And it's doing that in different ways because I've written the prompt slightly differently in both options. And you can see just the variety of output and the fidelity of output that it creates. Another feature that brands really love is our image mixer. So this is really unique in the AI space because most AI image mixers, they won't allow you to hold the silhouette. And that's really important for our clients as they're thinking about carryover styles or refresh styles. For example, here, we're starting with these white pants. But we want to see what these pants could look like in an ombre rainbow fabric. And you just simply upload that fabric. And you can scale it as well to the scale that you want. And it will really just render it as photorealistic for you. You don't have to get a sample. You don't have to, you know, go through a whole product photography process. And then as Kunal was mentioning, for prints, Raspberry is unique in that all of the prints that are generated through Raspberry are guaranteed fully seamless repeats. They are one-way repeats currently, and we're working on two-way repeats. And they're all custom. They're all license-free. They're all royalty-free. So we've saved a lot of our clients quite a bit of money purchasing prints from other suppliers or for their in-house graphic print artwork. It's just a massive time saver. Kunal, I don't know if you want to say anything else here.

So for those of us who have art studios and create prints, either for your own businesses or for retailers, this is a massive amount of time that goes into— let's say you purchase a piece of original artwork from a studio which is in repeat and is separated, it still needs some cleaning up, there's still scalability that needs to be looked at. With Raspberry, it really generates it very, very quickly and you can customize it even after you've gotten the final iteration done or use the print as is. But the most surprising, pleasantly surprising feature for our users was the fact that it came separated and in repeat. We have a few design teams that didn't have access to a dedicated textile artist, and this has really helped a lot. The other aspect of the whole print segment is also the use of Raspberry and graphics. As some of us know, Shutterstock is a widely used, you know, service that you tap into for screen tees or graphics, and there's problems with using a lot of those images because of, you know, indemnification if you're using the free service. But with Raspberry and the ability to modify a screen and really prompt it to change it, that takes everything out of the equation in terms of, you know, any issues with indemnification. And we have a team specifically that is going to market in the next month with all of the screens generated in Raspberry. So very useful to start either with a reference image or with the prompts and arrive at the desired result. Thank you.

All right. So we're going to blast through some of the other features here. So one is colorways. I know this is very important to a lot of brands who serve, you know, hundreds of colorways per season. So in Raspberry, you can actually batch recolor based on your current ASC palette or your current color story. And we just announced earlier today we have an integration with Coloro. So you'll be able to use Coloro codes in Raspberry and working on other integrations with other color providers soon. Um, on the editing side, I know it's really important in AI to have control, and that's something that's missing from a lot of platforms. For us, the way you can get control is through our edit module where you can brush over just like you would in Photoshop. You know, we have lasso tools, we have polygonal tools, we have brush masks. You can brush over specific areas. So for example, here we're brushing over her arms so that we can add long sleeves to her dress and nothing else about this image will change. We also have a technical drawings feature which allows you to save some time on the technical CAD sketches. So if you have a product photo, maybe you've lost your flat sketch somewhere, it's just in your massive folder, you can just really quickly generate one with Raspberry and it's compatible with Adobe Illustrator and is in a vector format to save you some time. And then I know a lot of folks here must be leveraging 3D. So one of the, the ways that AI and 3D cross together is our PhotoReal 3D to photo realism feature, which takes your CLO avatar, Browzwear avatar, DAZ avatar, and really, you know, enhances their look and feel. So enhances their skin, enhances their facial features. I know this is very popular amongst like kidswear designers because it just really brings that character to life. And now I'm going to go into some of the more ecommerce, marketing creative, and goals. So these are new features we launched in April of this year. Very similar to the design process, we are trying to really help save time and save cost. Here, instead of in the design process, we're helping to reduce the time it takes for photoshoots, allow you to maximize content and augment your photoshoot content. Because you can only shoot so many things in a day, right, or 2 days, and also help you reduce your photo cost, photo shoot cost. The first thing I want to talk about is diversity. So 85% of customers feel that diversity is very important, and they want to feel represented when they make purchase decisions. And AI is a good way to think about how we could achieve that, whether that's for internal presentations or externally facing. These are all AI-generated. This is also an AI-generated image. The one on the left is actually a real photo shoot that one of our clients did. And for them, they are based in Asia. It's very, very difficult for them to find models of other ethnicities. And so we're actually able to take the exact same photo shoot and just use that edit module to brush over the face and brush over the hands and in any skin again to change the models while keeping the garment integrity. And in terms of ROI, so we're really just comparing ourselves to a traditional photo shoot, right? I don't think when anyone starts exploring AI-generated photo shoots, they're thinking, oh, I'm going to eliminate 100% of my traditional photo shoots. They're going to think very strategically about, OK, What am I not able to do today with my traditional photoshoot budget that I want to be able to do to increase my conversion or products I'm not able to shoot and I'm still posting the CADs on New Order a Drawer? And what are items that we want to use AI to leverage? And what are the editorial campaigns we still want to shoot with, you know, the real crew and everyone? So I think for AI, I think obviously that the cost is 2+ orders of magnitude cheaper. The time is also an order of magnitude shorter. Even with factoring in, you still may have to do some retouches with your AI photo shoots. It's still so much faster and it's just really flexible as well. And there's no sustainability impact in terms of having to fly a crew of 30 people to Saint-Tropez and have that shoot in Saint-Tropez. Anthropée. So marketing features we want to cover. First one is the virtual try-on or on-body visualization. This really allows you to take the garments you might have designed in Raspberry, your teams have designed in Raspberry, and then automatically put them on a model. So you're really seeing like how it, how it will drape, how it will look and feel, and One, your designers can tell their story and vision better. And then two, your marketing team can actually start to use this for consumer testing. Some of our brands, they're leveraging this to test with their consumers before deciding how much to buy and manufacture. And then taking that asset one step further, you can see it's the same exact shirt, but we're actually able to generate multi-view. So here we started We started with a front view, and now we're asking it to generate a left, a right, a back. And you can determine the shot length as well, if you want it close up or a little bit zoomed out. And it holds everything— her, her hair, everything she's wearing— it holds consistent. And we've seen brands who increase the number of multi-view shots on their website reach double-digit conversion. Increases as well. That's kind of like the standard PDP. But I know we're also excited about action shots, especially sportswear brands. And I wanted to show Naomi Osaka again with that ombre red render we made previously. We're going to put it on. She's about to hit a massive forehand. And we're going to put her at Roland Garros. And yeah, I mean, these again, all AI generated and I think just really communicates what that end vision is. Lastly, I'll end on the Ability for AI marketing campaign. So this is more like your editorial style, right? You know, you're, we're on a yacht, we're in a runway show with graffiti, we're in Saint-Tropez in a pool. And all these are AI generated in Raspberry. Some of them have products that are real using OnBody and some of them are also just generated products. They don't exist. And then you can also generate backgrounds as well. So these are real bags from Balmain and the backgrounds, the kind of like white Western backgrounds are all AI generated. Um, I'll end on a slightly boring note, but important note as you're, as you're thinking about your AI adoption, which is security, personalization, and IP rights. Um, so we've worked with a number of IT and security teams on making sure that Raspberry is a safe solution for the teams to embark on. Um, and I think the differences with Raspberry is fourfold. Number one, uh, we can fully customize to your brand's heritage and aesthetic. So this is really important for brands who have a very strong brand aesthetic and look and feel. Um, secondly, we are SOC 2 compliant, so everything is fully secure. Your data coming in, your data going out, it's not being used by us to train the models further. Um, it's really your data and you own all the IP that you generate. And we do train all of our models to better fit the fashion industry. We have our own fashion designers in-house who actually label every single image that goes into our model training. And with that said, we're at booth 22. So please come visit us if you have any more questions or want to see more detailed demos. We're going to show some of that in our booth. And there should be a slide for questions, but also feel free to just chime in with questions. Thank you guys so much.

Okay.

Hello?

Okay.

Oh, it's working again. Question: The successes are apparent by utilizing these tools. What do you feel is the biggest hurdle stopping brands from fully adopting? Very good question. I think what we've noticed in the brands that have really done their adoption process quickly and well is there has to be leadership on their side, whether that's executive sponsors. Very important to have like DPC leaders or folks who are very familiar with technology and driving change management. Driving maybe their past 3D transformation, helping to make sure everything is moving along. I would say a lot of brands that are not as successful in AI, they just have the approach of, oh, we'll just play with the tool. But I think playing with the tool, it doesn't work as well as having a clear plan of what are the business goals we want to achieve? What are the things we can't do right now that AI can help us do and really have the executive sponsorship and executional sponsorship to make it happen. Second question, to train AI, it takes heavy lifting. Agree with this. How are team globally managing the ethics and energy consumption needed to achieve such successful results? So I think for us, I mean, we're certainly not a foundational model company. We're not OpenAI or Google AI. AI or Anthropic AI where they do massive amounts of training. We are really just trying to fine-tune our models to fit the fashion space and learn terminology like Gore-Tex better. So it's not as heavy on our side as it is for a lot of other big companies. Are the print patterns owned IP by the client? Yes. They are— everything you generate in Raspberry would be owned by the client and everything is unique. So if you and I are putting in exactly the same prompt at exactly the same time, we'll actually both get completely different outputs. Can Raspberry generate a tech pack based off of images generated that pulls internal blocks or specs from PLM to make first sample more accurate for fit? Very good question. Currently we cannot. Generate blocks or patterns. That is one of the most requested features in Raspberry, of course, and so we are trying to figure out how to do that. But AI is not the smartest in terms of measurements. So we're not quite there yet, and I don't think I'm aware of any AI companies that are there yet. What's the resolution of output from Raspberry? So all the prints are 300 DPI. And then any other— anything else you generate in Raspberry, you can scale up to 2K or 4K resolution. Working with ColorO, does the color data used come from the reflectance data to display the color accurately under all light sources, or is it just one light? Again, very good question. So we are using ColorO to the best of our ability. That being said, we're not You know, it's not the color chip, physical color chip. So I'm not sure if it's working accurately under all light sources. Can virtual try-on module use 3D renders? Yes. So you— we have a lot of clients who actually, they've built all their 3D avatars and then they use the 3D avatar to photoreal to make him look pretty. And then they use that asset in the virtual try-on to put new clothes on that avatar. When designs change throughout the development— oh, these are questions for Kunal. Which one is that?

The first two. The first two. Is sketch-to-render output 100% accurate for what design intends to see? Accuracy in the physical, or is this tool used to get close? I would say it's used to get close. If you're talking about accuracy that you can achieve with a traditional 3D software such as CLO or Browzware, where you capture the physics, you capture the actual fit intent, I would say that is not what you're going to get here, but it's a good starting step. It's a good way to visualize what your final product is going to end up looking like. And then you can proceed into the 3D workflow. So when designs change throughout the development, does Kunal's team update the images in Raspberry to maintain one-to-one accuracy? You can. Again, it depends on team to team. What's unique about Kelwood is we have multiple design teams working in different workflows. As I mentioned at the beginning, it's whatever works for each individual team. So it's really up to you as to what you want to maintain. Thank you.