Artificial Intelligence Draws Ink Wash Painting Based on the Data Generated by Kyung Hee University
Professor Hyoung-min Na of the College of Fine Arts is compiling the production data on Korean traditional-style ink wash painting as part of the 2023 Artificial Intelligence Learning Data Construction Support Project supported by the Ministry of Science and ICT and the National Information Society Agency
Professor Hyoung-min Na of the College of Fine Arts is working on the creation of Korean traditional ink wash painting production data as part of the 2023 Artificial Intelligence Learning Data Construction Support Project supported by the Ministry of Science and ICT and the National Information Society Agency (NIA). This is the second component of the project that Professor Na has been working on since last year in the category of Korean traditional ink wash painting production data by style in cooperation with ALLBIGDAT Co., Ltd. and Korea Deep Learning Co., Ltd., IT venture startups originated in Kyung Hee. I met Professor Na, who has passed the turning point of the project and heard about this project. <Editor's note>
Production of ink and color wash painting data; construction of text-to-image generation module
Q. First, I would like to hear your thoughts on being selected again for the project.
We have been thankfully chosen again for the same project following last year’s, which I find quite meaningful in that our previous work was well-received. We will also work together once again with ALLBIGDAT Co., Ltd. and Korea Deep Learning Co., Ltd., our partners from last year. After completing last year’s work, our progress rate was the best among all participants of the project, and we received the citation of excellence as a result. I am so proud of this achievement, especially in the sense that it was achieved through the combined efforts of Kyung Hee community members.
Q. What is the difference of this year’s project from last year’s?
The key difference is the material. Last year’s project created a conversion and creation model by producing data for each style of Korean traditional ink wash painting. This new project aims to build a text-to-image AI creation engine based on Korean ink and color wash paintings. Starting in early August 2023, arts students from Kyung Hee and other leading universities created over 4,200 ink and color wash paintings, which were then digitized, text-labeled, and learned by artificial intelligence. The end result is that, when you enter a text-based command, a corresponding image in the style of Korean traditional ink wash painting can now be generated.
There already are many AI engines that can create images by text-based command, and many arts students use these AI tools creating their own works. But most of these AI image generation engines can only produce images that are predominantly in western painting style. This problem stems from the type of basic painting data pool that had been fed into the AI modules, leaving only a small number of AI-generated examples of Korean painting styles such as ink and color wash painting. Even the few that are currently available are largely of mediocre quality compared to the level of sophisticated AI-generated images in Western painting styles.
The students participating in this project are selected from eleven fine arts colleges in the country, including our College of Fine Arts, who were handpicked for their advanced level of skill and talent in Korean ink and color wash paintings. The entire talent pool numbers about 180 painters including both student and professional artists, so I have a great deal of confidence in the quality of the raw data we would have. This pioneering project will be a milestone in the field of AI-based image generation engine based on Korean traditional-style ink wash painting.
Q. I am curious about the work creation process, which is the basic step that determines the quality of data.
Last year's project focused on producing data for each subset painting style variation within the stylistic family of Korean traditional ink wash paintings. In this year’s project, we are focusing on objects drawn in a specific style. Broadly speaking, they are artificial objects and natural objects, and we decided to follow a system of representative classification that artificial intelligence can understand and learn. A variety of objects, divided into a total of twenty major categories and 74 subcategories, are designed in accordance with the painting styles (Ilpil [drawing with one stroke] or Gongpil [drawing with numerous fine strokes]), painting methods (Molgol style [no outline style] or Guleug style [outlining style]), brush stroke techniques (Cheolseonmyo [consistent stroke] or Nanyeopmyo [orchid stroke]), and so on.
The painters’ skill level was assessed first, and then they were asked to choose a field that suited them best out of 74 subcategories. After they completed their work, professional instructor-level team leaders of the project group evaluated the painting. If the quality of the drawing was deemed unsatisfactory, it was redrawn. Only the drawings that had passed the filtering process were used for artificial intelligence learning. Afterwards, we plan to go through a second inspection process with professors of Korean traditional painting. Through the rigorous quality assurance process, we strive to improve the objectivity of evaluation and the qualitative completeness of the final product.
Q. Could you tell us about the post-production process?
After creating a suitable painting, digitizing the work is a process of refining the image. Afterwards, a text message describing the image in both Korean and English is prepared, which is at least 20 syllables or five words long in each version to ensure that it contains a sufficient level of information for AI to learn. The text label is written in Korean first and translated into English, which then goes through the QC process to check for error before being fed into the AI module.
Fostering student exchange and changing their perception on artificial intelligence in the field of fine arts
Q. The results of the project are expected to manifest in both education and research. What results are there?
As we can see from the Korean word for arts (“미술[美術]”), technology is a key component of artistic creativity. I believe now is the time for a new form of painting style to emerge that goes beyond the traditional medium of canvas, brush, and paint, which is knocking on the door to a completely new world of artistic expression led by artificial intelligence. AI will undoubtedly play a significant role in the future of art, which reminds me of the shock in the early modern era when the camera was first invented. For the painters at the time the shockwave caused by the invention of photography was a sense of existential crisis that threatened nothing less than “the extinction of painting.” However, photography and the art of painting eventually learned to coexist in related but separate spheres. Through this project, we expect to enter an era where new forms of art executed by artificial intelligence will unfold. I am proud that our College of Fine Arts is playing a central role in this process.
The educational aspect is also a significant bonus. As we gathered incredibly talented students from all major art colleges, it became a major opportunity to see and interact with other people's work. While they might once have been competitors at the time of the university entrance exam, they are now colleagues. Students participating in this project will work for a period of about three months, which would essentially be a big art convention that they are paid to attend, work together, engage, and learn. After 10 or 15 years from now, the students who worked together in this project may call themselves the “Seolleung school of painting” as they fondly reminisce on their memories.
There is a free flow of ideas while refining the technique of your major skills, which often leads to a significant change in one’s perception not only on paintings but also on traditions, as the new insight on evolving technology and how it would impact the future is key to understanding what lies ahead. For painters, for instance, complete immersion in the creative process might slowly force them to concentrate ever so narrowly on specific aspects, which frequently results in an intractable deadlock. But expanding the horizon of one’s perception might shed light on another way around the dead end; for instance, a myopic impasse on the technical aspect of painting might easily be resolved with the help of new digital technology and AI. When I think about the possibility of opening up a new avenue for future painters, I feel a profound sense of duty as the person overseeing the project.
Q. “Expanding the horizon of perception” is impressive. Is this from your personal experience?
In fact, prior to this project, I personally had little interest in artificial intelligence. Since I majored in Korean traditional painting, I focused solely on wielding brush strokes as a fine art painter. Integrating digital technology into this process? That never entered my mind at all. Last year, however, when I participated in the project for the first time, I realized that I must break away from the conventional concept of art. I believe that if the members of Kyung Hee community have an open mind, we will have the opportunity to be more innovative, ultra-modern, and lead the change of the times.
The importance of fine arts created by our hands still remains, as the high quality of the artwork as raw data is fundamental for artificial intelligence to produce excellent results. To this end, the foundational capabilities of pure art must be constantly cultivated and maintained. To summarize, I came away convinced that I must focus as much as possible on my main area of expertise as a Korean traditional painter while always having an open mind and keen interest for new technology.
Q. Is there anything you struggled with or focused on during the convergence process?
Convergence is a major topic these days not only at Kyung Hee but also in broader academic and societal contexts. However, there was a great concern on whether the academic environment afforded by university was conducive to fostering convergence. For my part, it was clear that as much of an expert painter as I was, I was completely ignorant of anything related to artificial intelligence. Therefore, there is a great need to collaborate with researchers in the fields of computer and software engineering. While Kyung Hee as a comprehensive university has both fine arts and computer science departments, there is a need for a natural atmosphere of horizontal convergence leading to free communication and engagement among professors and researchers from diverse fields. I believe this type of grassroot flow of natural horizontal convergence is much more desirable and effective than top-down or government-driven attempts.
Change is also urgently needed from an institutional perspective. I understand that related departments are working to improve and develop an infrastructure. There are also many national projects that reflect a changing society. For universities to effectively undertake such projects, a system must be in place to foster and incentivize innovation. While NIA may not provide direct benefits for the industry-academia cooperation group due to the absence of indirect cost subsidy, the agency has actively supported our project from its initial stage. I hope that this support and interest will serve as the catalyst for undertaking nationally significant initiatives leading to more fruitful results in the interconnected research fields.
- University Communication & Press