Inspired by AI? A Novel Generative AI System To Assist Conceptual Automotive Design

Ye Wang, Nicole B. Damen, Thomas Gale, Voho Seo, Hooman ShayaniarXiv 2024 · OpenReview

Abstract

Design inspiration is crucial for establishing the direction of a design as well as evoking feelings and conveying meanings during the conceptual design process. Many practice designers use text-based searches on platforms like Pinterest to gather image ideas, followed by sketching on paper or using digital tools to develop concepts. Emerging generative AI techniques, such as diffusion models, offer a promising avenue to streamline these processes by swiftly generating design concepts based on text and image inspiration inputs, subsequently using the AI generated design concepts as fresh sources of inspiration for further concept development. However, applying these generative AI techniques directly within a design context has challenges. Firstly, generative AI tools may exhibit a bias towards particular styles, resulting in a lack of diversity of design outputs. Secondly, these tools may struggle to grasp the nuanced meanings of texts or images in a design context. Lastly, the lack of integration with established design processes within design teams can result in fragmented use scenarios. Focusing on these challenges, we conducted workshops, surveys, and data augmentation involving teams of experienced automotive designers to investigate their current practices in generating concepts inspired by texts and images, as well as their preferred interaction modes for generative AI systems to support the concept generation workflow. Finally, we developed a novel generative AI system based on diffusion models to assist conceptual automotive design.

Introduction

In the professional automotive design studio, the conceptual phase is a delicate balance of creative exploration and technical constraint. While generic generative AI tools like Midjourney or Stable Diffusion have demonstrated remarkable capabilities, their application in professional workflows is often hindered by style bias, a lack of industry-specific semantic nuance, and fragmentation from established collaborative processes. For instance, a "bold" wheel design in a professional context implies specific structural qualities that generic models often fail to capture. This research, a collaboration between Autodesk and Hyundai, addresses these gaps by developing a specialized generative AI system tailored for conceptual car wheel design.

Methodology

The development of the system followed a rigorous, human-centered research path to ensure alignment with professional needs. The process was divided into three primary phases: a text inspiration survey involving 47 designers to understand current practices, a data annotation task with 16 participants to map design keywords to visual datasets, and in-person workshops to test interaction modes and prototypes.

Overview of the multi-stage human-centered research methodology.

Through these stages, the researchers "taught" the system the visual language of automotive design, mapping subjective keywords like "agile" or "sturdy" to specific geometric features.

System Design and Implementation

The core of the system utilizes Versatile Diffusion, chosen for its ability to handle multiple semantic inputs simultaneously. To ensure the outputs are practically useful, the researchers implemented custom denoising sub-processes that enforce radial symmetry and circular masking, ensuring that generated concepts maintain the fundamental proportions of a car wheel.

A key innovation is the Hierarchical Interaction model. Unlike standard "black box" prompt interfaces, this system allows designers to group inspiration images under specific concept keywords and adjust their relative "weights." This mimics the way designers naturally organize mood boards and allows for fine-grained control over how different sources of inspiration influence the final output.

System interface showing the integration of initial sketches, concept images, and generated wheel designs.

Key Insights from Professional Designers

The research revealed several critical distinctions in how professionals view AI assistance:

  • "Inspired" vs. "Influenced": Designers seek inspiration from non-automotive sources (e.g., architecture or nature) to maintain originality. They view direct influence from existing automotive products as a negative trait that limits creativity.
  • Novelty Over Resolution: During the conceptual phase, designers prioritized "unexpected and interesting" low-resolution sketches over polished, high-resolution renderings that were perceived as too "safe" or conventional.
  • Serendipity in Generation: Rather than a single "Generate" button, designers preferred a continuous, infinite-scroll interface that facilitates the serendipitous discovery of new ideas.

Conclusion

By bridging the gap between advanced diffusion models and professional design workflows, this system demonstrates how AI can act as a "junior designer" partner in the ideation process. The integration of physical constraints like symmetry and the support for hierarchical, weighted prompting provides a level of control previously missing from generative tools. Future work will focus on expanding these capabilities to more complex automotive components, such as full exteriors and interiors, and improving the explainability of AI-driven design choices.

Summary updated

Abstract

Design inspiration is crucial for establishing the direction of a design as well as evoking feelings and conveying meanings during the conceptual design process. Many practice designers use text-based searches on platforms like Pinterest to gather image ideas, followed by sketching on paper or using digital tools to develop concepts. Emerging generative AI techniques, such as diffusion models, offer a promising avenue to streamline these processes by swiftly generating design concepts based on text and image inspiration inputs, subsequently using the AI generated design concepts as fresh sources of inspiration for further concept development. However, applying these generative AI techniques directly within a design context has challenges. Firstly, generative AI tools may exhibit a bias towards particular styles, resulting in a lack of diversity of design outputs. Secondly, these tools may struggle to grasp the nuanced meanings of texts or images in a design context. Lastly, the lack of integration with established design processes within design teams can result in fragmented use scenarios. Focusing on these challenges, we conducted workshops, surveys, and data augmentation involving teams of experienced automotive designers to investigate their current practices in generating concepts inspired by texts and images, as well as their preferred interaction modes for generative AI systems to support the concept generation workflow. Finally, we developed a novel generative AI system based on diffusion models to assist conceptual automotive design.
BibTeX
@misc{wang2024inspired,
  author = {Ye Wang and Nicole B. Damen and Thomas Gale and Voho Seo and Hooman Shayani},
  title = {{Inspired by AI? A Novel Generative AI System To Assist Conceptual Automotive Design}},
  year = {2024},
  eprint = {2407.11991},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2407.11991},
}
Figure