A Generative Deep Learning for Exploring Layout Variation on Visual Poster Design

Abstract

Layout variation is an essential concept in design and allows designers to create a sense of depth and complexity in their work. However, manually creating layout variations can be time-consuming and limit a designer's creativity. The use of generative art as a tool for creating visual poster designs that emphasize layout variety is explored in this study. Deep learning through generative art offers a solution by using an algorithm to generate layout variations automatically. This paper uses the VQGAN and CLIP approach to describe a generative art system, which renders images via a text prompt and produces a series of variations based on the zoom parameter 0.95 and shifts the y-axis 5 pixels. Our experiment shows that one frame can be generated roughly in 10.108±0.226 seconds, significantly faster than the conventional method for creating layouts on poster design. The model achieved a good quality image, scoring 4.248 using an inception score evaluation. The layout variations can be used as a basis for poster design visuals, allowing designers to explore different visual representations of layouts. This paper demonstrates the potential of generative art to explore layout variation in visual design, offering designers a new approach to creating dynamic and engaging visual designs.