Investigating Prompt Transparency in Text-to-Image Systems

Tasks

• Literature review at the intersection of HCI and generative AI in creative fields with a focus on transparency and explainability
• Build an interactive system aiming for prompt transparency in generative AI. Implement SHAP and Integrated Gradients on text inputs conditioned by the output images to detect omitted elements. Implement Visual Question Answering (VQA) and Segmentation to detect unsolicitedly added elements in the image
• Evaluate the prototype in a user study with creative practitioners (illustrators)
• Document the system and findings in a Master Thesis and present the results in the Disputationsseminar
• (optional) Co-write a research paper based on the results