Towards an Automatic Prompt Optimization Framework for AI Image Generation.

Generative AI experienced a boom in 2022 with highlights such as the releases of Stable Diffusion for image generation and ChatGPT for conversational text generation. Although millions of images have been generated using products such as DALL-E 2, DreamStudio, and Midjourney, the learning curve for developing good text prompts that can lead to high-quality images remains steep, especially for inexperienced and less-technical users. Although various prompt engineering guides and tutorials have been developed to provide tips and guidance on prompt writing, there has been scant research on automatic prompt improvement algorithms and methods. In this paper, we present an automatic prompt optimization framework based on NLP analysis of a large prompt database and various machine learning models. A product based on our framework was developed and deployed for two months and real data were collected to evaluate our framework.