Design for Diverse Body Shapes with AI
Exploring AIGC Applications for DINED Mannequin Generation
Y. Ma (TU Delft - Industrial Design Engineering)
Toon Huysmans – Mentor (TU Delft - Human Factors)
Derek Lomas – Mentor (TU Delft - Form and Experience)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
What
In this project, a workflow method integrating generative AI has been developed. This versatile approach can be applied to a variety of body shape-related design tasks. For instance, designers can leverage this method effectively to generate design inspirations, enabling them to establish detailed design themes and goals more efficiently. Building upon this workflow, a new tool named DINED AI has been specifically designed to offer a more user-friendly and guided approach to accomplish these tasks. Although DINED AI is currently in the form of functional prototypes, both the user interface and back-end technology have been individually crafted to provide a realistic demonstration of its capabilities.
Why
With the rapid advancements in large language models, integrating AI technologies has emerged as the prevailing trend in product development. The aim of this project is to explore the potential of creating an AI-powered platform for DINED. DINED is renowned for its specialized expertise in anthropometry, which allows for the generation of realistic data models to aid in the design of ergonomic products. Leveraging this distinctive capability of DINED, and under the guidance of the mentors, this project focuses on the design of clothing, a widely recognized product that is intimately connected to the human body shape's form. Additionally, a streamlined workflow method will be developed to facilitate the design process.
How
The project is centered around leveraging technological advancements by implementing Stable Diffusion, LoRA, and ControlNet as crucial components. Stable Diffusion is a well-established text-to-image algorithm, while the LoRA model effectively refines Stable Diffusion using a limited number of images in the training dataset. On the other hand, ControlNet possesses the unique capability to extract specific control elements from the 3D mannequin generated by DINED, thereby enabling enhanced control over the resulting images.
In order to ensure a seamless user experience, the project integrates these aforementioned technologies into a cohesive and streamlined workflow. Each step of the process is accompanied by comprehensive guidance, empowering users to effortlessly navigate the system and fostering a coherent user experience. Furthermore, the project underwent evaluation through expert interviews, further validating its effectiveness and potential impact.