The gap between creative ideation and technical production has long slowed the fashion industry. Traditionally, a design moves from sketch to technical flat patterns to physical prototypes, with each handoff requiring translation between formats that do not share structure — a process that typically consumes multiple sample iterations before a garment is approved. Style3D, a company specializing in 3D and AI technologies for fashion, is working to close that gap by connecting generative AI with structured product data, creating a more unified workflow that runs from design through to production.

Central to this effort is GarmageNet, a framework developed by Style3D Research together with Zhejiang University, Shanghai Jiao Tong University, and Zhejiang Sci-Tech University. Published in ACM Transactions on Graphics and presented at SIGGRAPH Asia 2025, GarmageNet tackles a core challenge in fashion AI: bridging 2D patterns with 3D garment forms. Most generative image models produce visuals without the underlying construction data — seam placement, sewing relationships, fabric behavior — that production teams need, which is why eye-catching AI concept images rarely reach the factory floor unchanged. The framework captures pattern outlines, 3D geometry, and sewing information in a single unified representation, allowing inputs such as text, sketches, images, patterns, or point clouds to be converted into structured assets. It can reconstruct sewing relationships and initialize physical simulations, supporting design exploration, digital sampling, and editing.

According to the research, GarmageNet is supported by GarmageSet, a dataset of 14,801 professionally created garments. In evaluations, the system demonstrated high accuracy: GarmageJigsaw achieved 99.16% precision and 97.13% recall in sewing-point identification, while GarmageNet reached a 91.41% simulation-initialization success rate on 150 complex patterns, with an inference time of roughly eight seconds per garment under test conditions. A unified dataset of professionally patterned garments at this scale addresses a known bottleneck in the field, where research on garment construction has historically been limited by smaller, less diverse datasets.

The research reflects Style3D’s broader commitment to advancing fashion technology. The company has published more than 30 papers at major conferences including SIGGRAPH, CVPR, and NeurIPS, and holds 106 granted patents worldwide — among them 58 invention patents and 12 granted in Europe and the United States — with another 155 pending. Style3D also operates an overseas R&D center in Germany.

The practical effects of these technologies are already visible in the industry. German menswear brand OLYMP uses Style3D and Assyst to integrate 2D patterns, 3D simulation, visualization, and digital showrooms. According to the company, this integration has cut style development from weeks to days, reduced the number of physical samples, and improved communication across design, sales, and production. Fewer physical samples also carry a sustainability dimension that has become a growing priority for apparel brands, as sample production consumes materials, shipping, and labor that produce no sellable garment.

Style3D’s ambitions go beyond content generation. As CEO Eric Liu told Just Style, “The most important development is the agent. We use agents to link different skills.” By using AI agents to coordinate product data, 3D design, simulation, and production preparation, Style3D aims to build a seamless workflow in which human teams retain creative and operational control. This mirrors a wider shift across software industries, from single-purpose AI tools toward agent-based systems that orchestrate multi-step workflows. For fashion, this shift from isolated AI tools to integrated systems marks a notable step forward, pointing to greater efficiency, reduced waste, and faster time-to-market.

Source: Citybuzz