텍스트 기반 이미지 생성 인공지능의 원리와 현황 = Principles and Status of Text-to-Image Generation Artificial Intelligence

다국어 초록 (Multilingual Abstract)

This article grasps the principles of text-to-image generation models and examines key issues related to this new image generation method. The text-to-image generation model was created by combining the large language model and image generation model, which revolutionized natural language processing. The field of computer vision has set a turning point for a leap forward thanks to language understanding with pre-trained learning and the advanced image creation ability of diffusion models. As if competing, Open AI and Google developed new models and led the advancement of image generation capabilities with technological improvements. As a result, it is now possible to generate very realistic, high-quality images simply by entering text prompts. Additionally, elements that are not in the learning data can be generated, increasing the diversity of images. The success or failure of a text-to-image generation model depends on the fit and quality of the text and image, and it is important to create an appropriate prompt. Artificial intelligence automatically generates information, but appropriate human intervention is required in the text-to-image generation model.