PromptVis: Prompt-Based Interactive Visual Analysis Method for Text-to-Image Creation

It has become a research hotspot to help people write prompts for text to image creation. To tackle the limitations in existing work in prompt engineering, we propose PromptVis, an interactive visual analysis method for text to image generation, assisting users in evaluating and iteratively refining prompt to create images. First, the system parses the components of prompt phrases and provides suggestions for improving the prompts, such as recommendation-related prompts. Second, the user input and system-recommended prompts are visualized as clusters to support interactive exploration. Third, the system automatically evaluates the prompts and generates images from multiple dimensions, providing guidance for prompt improvement. Fourth, based on the recommended prompts, the system adjusts the local content on the existing images, enabling users to preview the effects of prompt modifications. Through creating prompts by diverse users for case studies and questionnaire evaluation show that this method can effectively assist users in creating prompts.