Tracing Edits: Interactive Exploration of Conversational Image Editing Histories

Problem Statement

Generative image editing tools enable users to transform visual content through natural language instructions, yet most existing systems offer little insight into how these changes unfold. Users often cannot see what specific parts of an image were modified, how edits accumulate over time, or how different prompts shape the final result. This thesis investigates how interactive and explainable interfaces can make the AI-driven editing process more transparent and controllable. It explores ways for users to trace edit histories, revisit earlier versions, and understand the connection between textual prompts and visual outcomes. By examining interaction patterns in conversational image editing, the work aims to identify design principles that strengthen creative control, trust, and interpretability in human–AI collaboration, resulting in a prototype and evaluation that demonstrate how conversational and visual feedback can enhance transparency and usability in generative workflows.

Tasks

• Perform a literature review on conversational/generative image editing, explainability, and interaction design; identify key challenges and gaps in current approaches to edit traceability and user control.
• Design and implement an interactive web prototype that supports natural-language edits and intuitive exploration of edit histories.
• Experiment with visualization and interaction techniques that make changes understandable and navigable for users.
• Conduct a user study.
• Perform statistical and qualitative analyses of the study data.
• Write the Master’s thesis and present your findings in the Disputationsseminar.
• (Optional) Co-author a research paper submission to an HCI or AI venue.

Skills Required

• Solid programming skills in Python for prototyping and integrating AI-based components.
• Experience with web development frameworks such as React (frontend) and Flask or FastAPI (backend).
• Familiarity with machine learning libraries and exposure to conversational image editing models such as Gemini Nano Banana, GPT Image, FLUX, etc. is a plus.
• Knowledge of data visualization and interactive graphics using libraries or tools such as D3.js, Plotly, or Canvas/WebGL.
• Experience with RESTful APIs and integrating AI services or model endpoints into web applications.

References

• Talk-to-Edit: Fine-Grained Facial Editing via Dialog