Tapestry, the parent company behind Coach, has embraced generative AI through Adobe Firefly by training the program with the brand’s proprietary assets and creating a custom model.
Adobe Firefly Custom Models enable brands to create content that is unique to their campaign and brand style. By design, Firefly is strictly trained to exclude any asset that touches a brand’s proprietary IP. With Custom Models, brands can fine-tune the Firefly models by training them with their own IP.
A trend seen in the fashion industry in recent years is the concept of a digital twin. These are virtual replicas of physical products, such as outerwear or accessories and become popular as a way to enhance the product development cycle. Created through a mix of photoshoots and CAD technical drawings, teams could experiment with new concepts more easily while bringing in direct consumer feedback.
Training generative AI with a brand’s identity means that when text prompts are used to generate a new image with Firefly, the generation will match the brand’s identity.
For Tapestry, these virtual replicas have proven valuable not only for the design process but also across teams from digital marketing to consumer research.
The Tapestry team had a breakthrough moment when an initial test of their Firefly Custom Model produced tailored digital twins that accurately mirrored products already seen on store shelves. This provided the company with a way to scale a critical internal service—allowing teams to better ideate and deliver products that resonate with customers.
The Tapestry team gathered a set of images of existing Coach handbags and used them for training single-concept models for each bag, object, and style. It allowed Adobe Firefly to ingest different “Coach Codes,” which are design components such as the hang tag or nuances around materials and hardware. In the early test mentioned previously, the team used the prompt “tabby handbag made of shearling fluffy material” (the Coach Tabby bag, launched in 2019, has become one of the most popular handbags for the company). Firefly generated an image that stunned the team: the custom model was able to produce a virtual replica that looked nearly identical to a Tabby bag variation that was currently on the market.
Digital twins are widely used across teams at Tapestry. They are shown in focus groups to gather direct consumer feedback that informs the product. The Strategy and Global Visual Experience teams also use digital twins to scale content for everything from social media campaigns to in-store merchandising. And when their data science is enhanced with digital twins, teams can extract knowledge and insights that can be used to inform decisions and predictions on how they create, make, and sell products.
The rollout of Firefly Custom Models will enable new opportunities in the design process as well. Generative AI allows product designers to experiment with new concepts or to take advantage of cultural trends. And with customised models, designers can create variations while maintaining the core creative vision. They simply put their ideas into a text prompt and generate new images that become a starting point in the production process. It opens new creative avenues and allows the team to design at the speed of their imaginations.
“At Tapestry DPC, we understood the potential for generative AI to alleviate the bottlenecks that were preventing us from scaling our digital twins, but we needed a trusted partner that was reliable and ethical in their approach to the technology. We have a long relationship with Adobe, and our teams love their design tools. We felt confident in what Adobe Firefly could generate, and the result from our customised model will dramatically change how we support our brands from concept to consumer,” J.J. Camara, senior director of digital product creation, Tapestry said.
With the internal validation, Tapestry is now embracing Firefly Services as well—a collection of creative and generative APIs and services for enterprises—to augment other processes including automating the manual steps that are involved in creating images for internal merchandising portals.
