Home Technology Chapter 2 Diffusion models in product design: optimizing form toward digital prototyping
Chapter
Licensed
Unlicensed Requires Authentication

Chapter 2 Diffusion models in product design: optimizing form toward digital prototyping

  • Valerio Perna and Panagiotis Kyratsis
Become an author with De Gruyter Brill
CAD/CAM
This chapter is in the book CAD/CAM

Abstract

In the last 20 years, the use of diffusion models and text-to-image generators has covered a wide range of fields related to graphic design, interior design, architectural design, and product designproduct design, promoting the development of brand-new methodologies to tackle the different stages of design from ideation to prototypingprototyping through the implementation of intelligent AI-based tools. Such models have shown impressive results in inpainting, text-to-image, super-resolution, colorization, instance segmentation, and so on. The generic pipeline of their use involves a forward and reverse process to learn from existing data contained in a pre-determined dataset and a noise map-based sampling procedure to generate novel information from existing ones. Even though performing well in most domains, there are still some lacking points to be addressed when it comes to their professional application in complex topics such as product design. This chapter investigates the state-of-the-art efficiency of diffusion models in product design, emphasizing the challenges and opportunities associated with their application to 3D-oriented tasks. The study seeks to address the critical obstacles in employing diffusion models for 3D purposes, proposing a methodological pipeline that transition from 2D to 3D.

Abstract

In the last 20 years, the use of diffusion models and text-to-image generators has covered a wide range of fields related to graphic design, interior design, architectural design, and product designproduct design, promoting the development of brand-new methodologies to tackle the different stages of design from ideation to prototypingprototyping through the implementation of intelligent AI-based tools. Such models have shown impressive results in inpainting, text-to-image, super-resolution, colorization, instance segmentation, and so on. The generic pipeline of their use involves a forward and reverse process to learn from existing data contained in a pre-determined dataset and a noise map-based sampling procedure to generate novel information from existing ones. Even though performing well in most domains, there are still some lacking points to be addressed when it comes to their professional application in complex topics such as product design. This chapter investigates the state-of-the-art efficiency of diffusion models in product design, emphasizing the challenges and opportunities associated with their application to 3D-oriented tasks. The study seeks to address the critical obstacles in employing diffusion models for 3D purposes, proposing a methodological pipeline that transition from 2D to 3D.

Downloaded on 29.1.2026 from https://www.degruyterbrill.com/document/doi/10.1515/9783111585062-002/html?lang=en
Scroll to top button