Page 84 - AC-1-2

Arts & Communication Digital restoration with generative AI

intricate restoration of cultural treasures by meticulously conservators represents an indelible imperative, ensuring
identifying distinct roof tile variations. The pivotal essence the evolution of holistic and potent strategies within
of their undertaking stems from the adept employment of this domain. The forthcoming segment pivots toward an
AI in distinguishing nuanced deviations in cultural motifs, emergent paradigmatic trajectory: an intricate dissection
a vital prerequisite for precision-led restoration endeavors. of generative AI and its prospective significance in reviving
Parallelly, Li’s scholarship accentuates the virtuosity and restoring absent or impaired facets of cultural heritage.
[23]
of amalgamating virtual reality with AI in championing
the cause of cultural heritage conservation. This melding 3. Methodology
foregrounds the viability of embracing expansive, multi- 3.1. Selection of artwork and region of interest
dimensional strategies for heritage preservation.
For this investigative endeavor, the selected artwork was
A recent breakthrough in AI-centric scholarship Antoine François Callet’s (1741–1823) portrayal of Achilles
[24]
is epitomized by the work of D’Orazioet al. in 2023, Dragging Hector’s Body Past the Walls of Troy from the
wherein the trio delved into the potential of long short- years 1784–1785 (Figure 1). The artwork’s established
term memory neural networks. Their primary objective provenance is anchored in Pietro Antonio Martini’s
entailed processing maintenance requests from end-users, (1738–1797) panoramic engraving of the Salon from 1785
thus laying the groundwork for anticipatory conservation (Figure 2). A particular area of interest for this research is
paradigms. Such paradigms, if effectively deployed, could the top right section of the painting, showcasing a celestial
circumvent potential deterioration, thereby eliminating or demonic female entity, characterized by extended wings
the exigency for more invasive restoration techniques. This and armed with a likely dagger or athame. This specific
investigative trajectory finds resonance with the research segment was earmarked for experimental restoration,
endeavors of Moreno et al., where fuzzy logic served given the significant wear and tear surrounding this figure.
[25]
as the linchpin to evaluate environmental ramifications Notwithstanding the damage, the depicted figure remains
on heritage structures, thereby weaving together an largely preserved.
intricate tapestry of conservation strategies. Contrarily,
Bordoni et al. postulate that AI’s methodologies and 3.2. Software utilization and algorithmic approach
[26]
techniques stand as potent catalysts for the archival, The restoration experiment incorporated the use of the
conservation, and appreciation of cultural heritage. Their generative imaging platform known as Easy Diffusion
academic exploration offers an expansive vista, paving (referenced from Easy Diffusion README, n.d.). This
the way for successive research endeavors and shedding software is underpinned by the algorithmic framework of
light on the transformative potential of AI in the realm Stable Diffusion 2.1. The choice of this tool was motivated
of heritage conservation. Not to be overlooked is the by its economic viability and its adherence to the General
contribution of Ranaldi and Zanzotto, which ventures Public License (https://stability.ai/blog/stablediffusion2-
[27]
into the conceptual territory of self-empiricist logic. Their 1-release7-dec-2022). The software offers two primary
proposition articulates the transformative capability of modes of operation: text-to-image and image-to-image,
contemporary AI mechanisms in reshaping the traditional thereby allowing a range of flexibility for the restoration
paradigms of heritage preservation. process. Of particular note is the software’s “inpainting”
Emerging from the academic discourse is two feature, where users can selectively mask areas for flaw
pronounced trajectories: the first emphasizes the correction or detail generation.
conservation and rejuvenation of physical artifacts and
architectural edifices, while the second plumbs the depths 3.3. Initial recolorization and area-specific
of the ethereal dimensions of cultural heritage. Delving restoration
into the realm of the intangible, Yu et al. proffer an For the restoration, the experiment harnessed the
[28]
incisive exploration, accentuating the versatility of AI in capabilities of image-to-image prompts for inpainting, an
architecting recreational spaces dedicated to intangible approach grounded in the methodology prescribed by Oncu
cultural legacies. Regardless of whether the emphasis lies et al. The inaugural recolorization was concentrated on
[29]
in the adept utilization of neural networks for meticulous the painting’s background facets. Focal elements,—namely,
restoration endeavors or in the orchestration of foresight- the female figure, the brown fabric, the hands of Achilles,
driven models for anticipatory conservation, AI’s potential and the equine figure —benefited from a refined restoration
as a transformative agent in cultural heritage stewardship process, marked by the introduction of semi-transparent
stands uncontested. Consequently, the synthesis of color masks. The application of these masks was executed
expertise from AI connoisseurs with insights from heritage with precision, ensuring a congruent alignment with

Volume 1 Issue 2 (2023) 3 https://doi.org/10.36922/ac.1793