In the rapidly evolving landscape of fashion technology, Style3D is making a significant move to connect generative AI with structured product data, aiming to transform how fashion teams design, develop, and produce garments. The company's latest research, GarmageNet, presented at SIGGRAPH Asia 2025 and published in ACM Transactions on Graphics, addresses a core challenge in fashion AI: linking 2D patterns with 3D garment forms.
Developed in collaboration with Zhejiang University, Shanghai Jiao Tong University, and Zhejiang Sci-Tech University, GarmageNet introduces a unified representation that captures pattern outlines, 3D geometry, and sewing information. This framework allows users to input text, sketches, images, patterns, or point clouds, and generate structured assets, reconstruct sewing relationships, and initialize physical simulations. This capability is crucial for design exploration, digital sampling, and editing, marking a departure from AI tools that merely generate isolated images.
The importance of this development lies in its potential to integrate AI into the entire product development workflow. By linking inputs with patterns, sewing relationships, and simulation, GarmageNet exemplifies how fashion AI can become a functional part of a connected process, rather than remaining a novelty. This aligns with Style3D's broader vision, as CEO Eric Liu explained to Just Style, "The most important development is the agent. We use agents to link different skills." This agent-based approach is intended to coordinate product data, 3D design, simulation, content creation, and production preparation, while keeping human teams in creative and operational control.
GarmageNet is supported by GarmageSet, a dataset of 14,801 professionally created garments, which was used to validate the framework's performance. In evaluations, GarmageJigsaw, a component for sewing-point identification, achieved 99.16% precision and 97.13% recall. GarmageNet itself recorded a 91.41% simulation-initialization success rate on 150 complex patterns, with an inference time of approximately eight seconds per garment under test conditions. These metrics suggest that the technology is not only innovative but also practical for real-world applications.
Style3D's commitment to research and development is evident in its substantial R&D organization and its publication record, with more than 30 papers at prestigious conferences such as SIGGRAPH, SIGGRAPH Asia, CVPR, NeurIPS, and 3DV. As of 28 July 2026, the company holds 106 granted patents worldwide, including 58 invention patents and 12 granted in Europe and the United States, with an additional 155 patent applications pending. This intellectual property portfolio underscores the depth of its technological foundation in deformable-object simulation, garment deformation, digital fabrics, and AI-assisted workflows.
The practical implications of Style3D's advancements are already visible in the industry. German menswear brand OLYMP utilizes Style3D and Assyst to connect 2D patterns, 3D simulation, visualization, and digital showrooms. The company reports that this integration has reduced style development time from weeks to days, decreased the need for physical samples, and improved communication across design, sales, and production teams.
By bridging the gap between generative AI and structured product data, Style3D is not only enhancing design capabilities but also streamlining the entire fashion product lifecycle. As the industry moves toward digitalization, such connected workflows are likely to become essential, positioning Style3D at the forefront of this transformation.
