Image Time Machine: AI-Powered Transformation of Contemporary into Historical Photographs

Abstract

This research introduces a tool that utilizes artificial intelligence to transform modern
photographs into authentic representations of historical aesthetics. The tool integrates
advanced AI technologies, including Large Language Models (LLMs), Stable Diffusion
XL, ControlNet, and decade-specific Low-Rank Adaptation (LoRA) models, combined
with automated prompt engineering. The objective is to capture the unique visual characteristics of different eras, mimicking the imperfections of analog and vintage photography.
LoRA models are trained on historical images to enhance the authenticity of generated photographs. The resulting images were evaluated in a user study, which confirmed
that the LoRA models significantly improve the quality, realism, and historical accuracy compared to images generated without them. Future enhancements may involve
refining LoRA models and expanding to more historical periods. This research uniquely
intersects AI, photography, and historical aesthetics, offering new ways to experience
the present through AI-generated vintage imagery.

Date of Award	2024
Original language	English (American)
Supervisor	David Christian Schedl (Supervisor)