Visualization for a new era: Impact and application of large language models and AIGC to traditional business models

Abstract

<p>This paper focuses on the application and business value of large-scale language models, such as GPT and Ernie’s model. These models combined with AIGC tools like stable diffusion generate images with fixed styles, character traits, and continuous plots using randomized story scripts. As a result, it enhances the operational efficiency between or within industries widely, and it fully demonstrate their business value. On the technical side, this paper describes in detail of building a pipeline to generate cue words required for stable diffusion, in which using large-scale language models and story scripts. Subsequently, the limitations of text-to-image are summarized by comparing the traditional method and language model, i.e. comparing characteristics from traditional book production and images generated using language model’s cue words. This leads to a supervised multiround iterative LoRA modeling scheme that utilizes CLIP to achieve character IP fixation. To evaluate the impact of the application direction, we combine application scenarios and researches on application aspects regarding current AIGC industry structure, we found that the AIGC tool has several major aspects, mainly includes the aspects of basic big model, industry and scenario models, business and domain small models, AI infrastructure and AIGC supporting services. big model and AIGC techniques generate images with no specific rules and have less limitation. We call this ‘visualization’ in the new AI era. In this paper, we explore the possible impacts and economic values when changing from traditional domain to the new AI ear.</p>