Research Article
AI in Art: Simulating the Human Painting Process
@INPROCEEDINGS{10.1007/978-3-030-95531-1_20, author={Alexander Leiser and Tim Schlippe}, title={AI in Art: Simulating the Human Painting Process}, proceedings={ArtsIT, Interactivity and Game Creation. Creative Heritage. New Perspectives from Media Arts and Artificial Intelligence. 10th EAI International Conference, ArtsIT 2021, Virtual Event, December 2-3, 2021, Proceedings}, proceedings_a={ARTSIT}, year={2022}, month={2}, keywords={AI art AI in Art Painting Semantic Segmentation Semantic Labeling Machine Learning}, doi={10.1007/978-3-030-95531-1_20} }- Alexander Leiser
Tim Schlippe
Year: 2022
AI in Art: Simulating the Human Painting Process
ARTSIT
Springer
DOI: 10.1007/978-3-030-95531-1_20
Abstract
While AI is being used more and more to generate images, the generation usually does not resemble a human painting process. However, for applications in the field of art, it is useful to simulate the human painting process—e.g. in relation to location, order, shape, color and contours of the areas being painted in each step. Such applications are for example when a robot paints a picture or a program teaches humans to paint. Consequently, in this paper we evaluate and compare different approaches to simulate the human painting process. Additionally, we present our solution for this task which is based on a combination of filters and semantic segmentation. In our survey, this approach was rated as better and more realistic than the most realistic approach for this task so far which is a reinforcement learning approach: In all surveyed categories—location, order, shape, color and contours of the areas being painted in each step—always a significant majority of the participants prefers our approach to simulate the human painting process. When we displayed two time-lapse videos with the painting process of Edvard Munch’s The Scream in parallel, even 79% found our generated process more realistic than the reinforcement learning-based process.
