Prompt Assisted Generative Modelling for Digital Recreation of the Lost Caesar Paintings by Titian
Ivan Rybnikov, Ivan Lysikov, Yan Antropov, Manuel Portela, Hassan Ugail, Maadh Zameer, Noah Hassan Ugail · 2024
This paper presents a machine learning approach to digital recreation of the lost Caesar paintings by Titian using prompt assisted generative modelling. We leverage Stable Diffusion XL (SDXL), a state-of-the-art generative model integrated with ControlNet and custom Low-Rank Adaptation (LoRA) models, to produce high-fidelity images that closely mimic the artistic style and historically accurate images resembling the originals. Our methodology emphasises the integration of domain specific knowledge, including detailed prompt generation and iterative refinements informed by art history experts, to ensure realistic and contextually appropriate output. The resulting images provide valuable insight into the appearance of these lost masterpieces, offering a unique perspective on Renaissance art. Additionally, this work highlights the potential of generative AI for the digital recreation of art and sets the stage for future exploration in combining AI with expert knowledge to preserve and revitalise cultural heritage.