A mesterséges intelligencia lehetőségei és korlátai a műtárgy-rekonstrukció területén. Töredékes és hiányos festett felületek virtuális kiegészítése mesterséges intelligencia segítségével és digitális rajzolás útján

Possibilities and limitations of the use of artificial intelligence in artefact reconstruction Virtual completion of fragmented and incomplete painted surfaces using artificial intelligence and digital drawing Generative artificial intelligence is capable of visualising our imagination and creating different kinds of illustrations based on prompts. These handy, mainly online applications have not only brought significant changes to graphic design but also affected conservation, retouching, and digital reconstruction. Several international projects are in progress or have already been completed in which researchers relied on the help of AI specialised for a certain artist. In this study, the Generative Fill feature of Adobe Photoshop has been used to create digital reconstructions of the selected paintings. This generative AI application is one of several developed for creative designers instead of conservators or restorers. Accordingly, these algorithms are not familiar with the proper art historical examples that could be used during the restoration process, and, therefore, their use in conservation has certain limitations. However, their great advantage is that they are now available to anyone, not just a narrow scientific circle. Two paintings on canvas, both with significant losses, have been selected for the current experiments. One is a coat of arms depiction in an 18th–19th-century fragmentary painting from Dunaföldvár, and the other is a Patrona Hungariae altarpiece from Nagykökényes by Joseph Karl Schöfft (1776–1851), on which the missing parts of the face of the angel kneeling before the Virgin have been reconstructed digitally. It has become clear during the current experiments that AI alone is not capable of creating professional and authentic reconstructions. However, whenever ethical questions arise about an actual restoration, AI can greatly shorten the working hours sentenced to make digital prefigurations and, with the countless image variations created in a short time, multiply our choices so that the best possible decision can be made from the painting’s point of view. Since the algorithm has access to an extensive visual base, it may generate a reconstruction version that did not occur to the restorer. An example of this was the face of the angel in the Schöfft altarpiece, where well-defined losses had to be dealt with. In cases where the losses affect more than half of the painted surface and are not concentrated, the tested AI version in its current stage of development has encountered problems. This happened in connection with the reconstruction of the coat of arms. The problems include the extent of the losses and the fact that AI was not trained to use and display the appropriate heraldic motifs. Until developing special AI applications trained in the style of a particular painter or period and heraldic depictions becomes a fast, simple, low-budget, and widely accessible option, we have to be satisfied with such extensions developed for general and creative purposes. In certain cases, these can also help the restoration process even in their current form.