Year: 2025
Type: conference-paper
Abstract: The current creative design field faces problems such as low design efficiency, lack of innovation, and difficulty in meeting personalized needs. To address these challenges, this paper proposes an intelligent creative design system based on a diffusion model. The system aims to improve design efficiency and innovation through intelligent technology while meeting the personalized needs of users. The system first uses a diffusion model to generate a variety of design sketches, gradually iterates the initial sketch through Gaussian noise, and introduces user-input keywords or style tags in combination with the conditional diffusion model to control the generation direction. Then, it uses a convolutional neural network (CNN) to extract features from the sketch, and uses a multi-layer perceptron (MLP) to optimize the lines and structure of the sketch. Then, it combines style transfer technology to transfer the user-specified artistic style to the sketch, and introduces an attention mechanism to ensure the consistency of local details in style transfer. Finally, it uses GAN to render the design with high quality, optimizes detail performance through adversarial training of the discriminator and generator, and uses super-resolution technology to improve image resolution. The system is 12.4% and 15.4% higher than GLIDE (5.47) and DALL-E 2 (5.33), respectively. The intelligent creative design system based on the diffusion model can effectively improve design efficiency and innovation, meet personalized needs, and provide new solutions for the field of creative design.
Authors: Siwen Sun, Changjiang Song, Lei Li, Yuwen Gao, Jingwen Yi, Yue Zhang
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4411539888