Generative Artificial Intelligence as a Catalyst for Change in Higher Education Art Study Programs

Abstract

Generative Artificial Intelligence (AI) has emerged as a transformative tool in art education, offering innovative avenues for creativity and learning. However, concerns persist among educators regarding the potential misuse of text-to-image generators as unethical shortcuts. This study explores how bachelor’s-level art students perceive and use generative AI in artistic composition. Ten art students participated in a lecture on composition principles and completed a practical composition task using both traditional methods and generative AI tools. Their interactions were observed, followed by the administration of a questionnaire capturing their reflections. Qualitative analysis of the data revealed that students recognize the potential of generative AI for ideation and conceptual development but find its limitations frustrating for executing nuanced artistic tasks. This study highlights the current utility of generative AI as an inspirational and conceptual mentor rather than a precise artistic tool, highlighting the need for structured training and a balanced integration of generative AI with traditional design methods. Future research should focus on larger participant samples, assess the evolving capabilities of generative AI tools, and explore their potential to teach fundamental art concepts effectively while addressing concerns about academic integrity. Enhancing the functionality of these tools could bridge gaps between creativity and pedagogy in art education.

1. Introduction

In recent years, artificial intelligence (AI) has undergone substantial advancements, leading to transformative innovations and broad applications of the technology [1]. The potential of AI in education is widely recognized, with international educational organizations emphasizing its capacity to revolutionize teaching, learning, and institutional operations [2]. AI-powered tools and systems, including generative artificial intelligence systems like GPT-4, Open Assistant, DALL-E, and MidJourney, have revolutionized higher education by offering data-driven decision-making, adaptive assessments, and personalized learning experiences [3,4]. These systems not only enable autonomous generation of human-like text and images but also engage in a variety of intellectual tasks, reshaping the traditional role of higher education institutions in knowledge transfer and skill development [1]. While these technologies offer innovative ways for students to explore and understand artistic principles, they also introduce challenges related to academic integrity and the authenticity of student work.

Tools like image generators and conversational AI can assist students in visualizing complex concepts, experimenting with different compositional techniques, and receiving immediate feedback. This technological integration can enhance creativity, foster independent learning, and prepare students for a digital-centric art industry. For instance, Hutson and Lang (2023) demonstrated that tools like DALL-E can enrich creativity in the classroom by offering immediate visual feedback, although they emphasized the importance of ensuring AI complements rather than replaces human creativity [5]. Similarly, Vartiainen and Tedre (2023) highlighted the reflective value generative AI brings to art education but noted concerns regarding algorithmic bias and copyright issues [6]. Despite these benefits, integrating AI tools into curricula presents challenges, particularly around ensuring students continue to develop critical creative skills.

Generative AI tools often struggle with executing precise artistic instructions, especially in tasks requiring nuanced understanding of composition principles. Research has shown the necessity for higher education institutions to adapt their curricula to nurture high-quality art design talent in response to the demands of an AI-driven era [5,7]. Furthermore, AI-enhanced learning environments have been shown to significantly improve student performance and motivation, as evidenced by Chiu et al. (2022), who reported that a deep learning-based art learning system enhanced student achievement and satisfaction in art education [8]. Generative AI may not yet be sophisticated enough to fully replace traditional learning methods for foundational art skills. Relying solely on these tools could impede the development of essential artistic abilities and critical thinking.

The integration of generative AI into art education is a burgeoning field, yet current research reveals significant gaps that warrant attention. Existing research often lacks a task-based analysis to determine the optimal use cases for AI tools within art education. Thus, this research addresses a critical gap by clarifying the specific artistic educational contexts in which generative AI tools offer the most substantial benefits or present notable limitations. Understanding the specific tasks where generative AI can enhance or impede the learning process is crucial for its effective integration. This involves identifying which aspects of art education, such as ideation, prototyping, or technical skill acquisition, benefit most from AI assistance, and where it might pose challenges, such as in fostering original creative thinking. Furthermore, this study highlights the importance of integrating AI literacy, particularly prompt engineering, into art education curricula to enable students to better align generative AI outputs with their artistic intentions.

Addressing these research gaps is vital for the informed and responsible incorporation of generative AI into art education. By conducting thorough investigations into the educational impacts of generative AI and performing detailed task-based analyses, educators can develop strategies that leverage AI’s strengths while mitigating its challenges. This study seeks to contribute to the growing body of knowledge by exploring how bachelor’s-level art students perceive and interact with generative AI tools in the artistic composition process, addressing the following research questions:

• What are the key challenges students face when using generative AI tools for artistic composition?
• How do students refine and iterate AI-generated compositions to align with their artistic intent?
• What are students’ attitudes toward the integration of generative AI tools in art education?

2. Generative AI in Higher Art Education

Generative AI, a subset of Artificial Intelligence, refers to technologies that utilize deep learning models to produce human-like content, such as text, images, and other media, in response to diverse prompts like languages, instructions, or questions. While generative AI technologies have been in development for some time, the release of tools like ChatGPT has significantly increased both public and academic interest in the field [9,10,11]. The evolution of AI itself can be traced back to Turing’s foundational work, which introduced the concept of integrating intelligent reasoning into machines [12,13].

The advent of generative AI platforms, such as ChatGPT, MidJourney, and Gemini, has led to transformative impacts across various sectors, including education, labor, and leisure [14]. User-friendly interfaces and widespread accessibility have contributed to the rapid adoption of generative AI tools [15]. For instance, ChatGPT reached one million users within five days of its launch and has since amassed millions of active users globally [16].

Generative AI systems operate by leveraging large language models (LLMs) trained on vast datasets. These models can produce text, images, audio, and videos often indistinguishable from human-created content [17]. Applications of generative AI in education include creating tailored learning materials, assisting with language translation, and generating creative outputs like essays and blog posts. According to Lim and colleagues, generative AI uses deep learning models to generate human-like responses to complex prompts, making it a valuable tool across various disciplines [9].

Generative AI has introduced innovative solutions for personalized learning, administrative support, and content generation in higher education. AI-driven personalized learning systems adapt to individual students’ psychological needs and contextual factors, improving knowledge retention and engagement [18]. Virtual tutors, for example, have been shown to increase student motivation and creativity by personalizing the learning experience [19,20].

Additionally, AI-powered grading systems offer scalable and efficient alternatives to traditional assessment methods, addressing challenges in large class settings [21]. Predictive analytics further enables institutions to identify at-risk students and design timely interventions to improve learning outcomes [22]. AI also streamlines administrative tasks, reducing faculty workload and enhancing teaching efficiency [23]. Examples include automating student admissions, financial aid processing, and library services [24]. Moreover, LLMs could revolutionize educational practices by serving as virtual instructors, curriculum developers, and contributors to scholarly work [11].

Generative AI enhances student learning by enabling tailored educational experiences, improving accessibility, and leveraging data-driven approaches to monitor progress. Tools like ChatGPT provide writing assistance, especially for non-native English speakers, fostering creativity and idea generation [25]. Text-to-image generators like DALL-E support teaching in the arts, enriching students’ technical and artistic understanding [26].

The integration of generative AI in higher education also raises ethical and practical challenges. Issues of plagiarism and academic integrity have become more prominent, as AI-generated content is often undetectable by standard plagiarism tools [27]. Additionally, biases in AI-generated content can reinforce stereotypes and marginalize certain groups; therefore, training datasets must be carefully examined [28]. The potential misuse of generative AI highlights the need for clear guidelines and ethical frameworks to govern its application in education [29]. Additionally, educators must balance AI assistance with fostering students’ critical thinking and independent learning skills [30].

Despite its limitations, generative AI offers transformative potential for education. Its ability to create personalized learning experiences, streamline administrative tasks, and enhance teaching efficiency makes it an invaluable tool for modern higher education.

Generative AI has demonstrated its capability to create poetry, visual art, and music, and even to choreograph dance movements [31]. These advancements necessitate understanding how individuals perceive AI-generated art and the personal attributes influencing these perceptions [32]. Previous research has concentrated on two primary questions: whether humans can differentiate between AI-generated and human-created art and whether biases exist toward AI-generated works.

Regarding the ability to discern between artworks, most studies suggest that individuals struggle to consistently distinguish human-made art from AI-generated art [33,34]. Similarly, a notable body of research highlights a negative bias toward AI-generated art [35], even in instances where the source of the artwork is unclear to participants. Notably, while the average viewer often categorizes AI-generated creations as “art” [36], debates persist about whether AI can genuinely fulfill the role of an artist. Critics argue that art inherently requires intention and communication, attributes that AI lacks independent of human input [37,38]. Consequently, AI-generated art remains reliant on human-designed algorithms, training datasets, and prompts, inherently tying its outputs to human involvement.

The integration of AI into artistic creation dates back to Harold Cohen’s AARON project in 1973, which translated visual decision-making processes into robotic painting systems [39]. Over subsequent decades, the evolution of machine learning, pattern recognition, generative adversarial networks (GANs), and text-to-image diffusion models has profoundly influenced artistic practices. These tools have empowered artists to explore new creative processes, challenging traditional boundaries of authorship and creativity [40]. Generative AI, by enabling the production of art-like outputs with minimal effort, has significantly democratized artistic creation, allowing a broader range of individuals to engage in creative expression. However, the blurred distinction between the artist and the tool raises critical questions about authorship, ownership, and creative agency [38]. While generative AI expands access to creative resources, it simultaneously redefines the relationship between human creators and their tools, necessitating further exploration of its implications for traditional artistic norms.

Concerns about AI replacing human creativity often underpin preferences for human-made art [41]. Such fears reflect deeper anxieties about preserving human identity as the sole source of creative innovation [42]. Moreover, ethical and moral considerations, including a desire to support human artists, may also influence evaluations of human versus AI-generated artworks. General attitudes toward AI vary across cultures and individuals, influencing its acceptance and integration into creative contexts [43]. For example, individuals with a favorable view of AI often attribute higher profundity and value to AI-generated creations [44]. Conversely, skepticism about AI’s role in creativity, particularly its potential to replace human effort, remains a significant barrier to broader acceptance [45]. The discourse surrounding generative AI in art emphasizes its role as a new medium rather than a threat to creativity. While it raises questions about authorship and originality, its potential to augment artistic practices and democratize access to creative tools presents opportunities for further exploration. As perceptions of AI-generated art continue to evolve, understanding the interplay between human and AI contributions will remain critical to navigating this transformative landscape.

Generative AI tools such as ChatGPT, DALL-E, and MidJourney have introduced novel opportunities for creative expression in educational contexts. These tools enable students to visualize abstract concepts, engage in collaborative creativity, and rapidly prototype ideas. Beyond enhancing artistic skills, generative AI supports the development of problem-solving and critical thinking abilities, particularly when educators are equipped to confidently integrate these technologies into the classroom. The adoption of generative AI in art education is reshaping traditional pedagogical approaches by augmenting creativity and facilitating personalized learning experiences. Research has demonstrated how generative AI aids fashion designers in ideation and prototyping, as seen in the development of tools like StoryDrawer for visual storytelling and AI-based garment design systems that integrate fashion expertise [46]. These applications highlight the synergy between human designers and AI, emphasizing the need for technical proficiency and creative adaptability in leveraging AI tools effectively.

Despite its potential, generative AI integration in art education faces several challenges, including ethical concerns about authorship, dependency risks, and accessibility barriers. Ethical concerns regarding intellectual property, over-reliance on automated outputs, and data bias emerge as critical issues for students who are still developing foundational artistic skills. For instance, Sullivan et al. (2023) [27] highlight the academic integrity dilemmas associated with AI-generated assignments, noting that traditional plagiarism checks are often insufficient when content is produced by generative models. Similarly, Dwivedi et al. (2023) [28] argue that reliance on AI can diminish learners’ opportunities to cultivate creativity and problem-solving abilities, which is a key consideration in art education, where hands-on experimentation is vital. Moreover, Tedre, Kahila, and Vartiainen (2023) [6] emphasize the importance of carefully structuring AI-based projects in craft and design classes to ensure that students learn critical thinking and prompt-engineering skills, rather than simply accepting AI outputs as final. Furthermore, the importance of “prompt engineering”, crafting precise inputs for desired AI outputs, has emerged as a critical skill in AI-assisted design, underscoring the necessity of specialized training for both educators and students. Taken together, these studies illustrate the need for educator training, transparent guidelines on AI use, and balanced curricular approaches to preserve students’ artistic autonomy while leveraging AI’s benefits for exploration and ideation.

Given the rapid development of AI technologies, it is inevitable that these tools will be increasingly integrated into creative industries, including art education. To fully understand how generative AI tools can be effectively utilized in these contexts, it is essential to explore how students engage with these technologies, the educational value they derive, and the challenges they present.

This study addresses the concerns of educators regarding the potential misuse of generative AI tools, such as text-to-image generators like DALL-E, which some view as unethical shortcuts that undermine traditional learning methods. The focus of the research is to examine how undergraduate art students use generative AI tools in their learning, particularly in a practical composition task, and to identify the educational benefits and limitations these tools offer. While the study highlights the potential of generative AI for ideation and conceptual development, it also emphasizes its limitations in producing refined, nuanced artistic outputs. By examining these dynamics, this study aims to bridge the gap in understanding how generative AI can contribute to art education. The following section details the methodology employed in this research, outlining the steps taken to observe student interactions, gather reflections, and analyze the data.

3. Materials and Methods

The research involved bachelor’s-level art students from a university’s art program. Although the class had 13 students, participation in the research was determined by attendance during the first lecture, where key instructions and study details were provided. Since only 10 students attended this session, they were invited to participate in the study. This ensured that all participants received the same information and began their tasks with consistent instructions. The participants’ ages ranged from 18 to 26, with a mix of male and female students. All participants had prior experience in traditional art practices, but their familiarity with generative AI tools varied. Some students had used AI for ideation and concept development, while others had little to no experience with AI tools.

This study met with ethical research standards, guaranteeing that involvement was entirely voluntary and had no bearing on students’ grades or academic achievement. The study’s goals and methods, and participants’ freedom to discontinue participation at any moment without facing impacts, were explained to them before the study began. We obtained informed consent from each participant. To protect participant identities, data were gathered and analyzed anonymously, and the study adhered to the Declaration of Helsinki’s ethical guidelines for research involving human subjects.

The study employed a mixed-methods approach, combining quantitative data from questionnaires with qualitative observations and participant feedback. The research was conducted in two stages over a one-month period:

• Week 1—Lecture on basic composition principles. Traditional task completion: students completed a composition task using traditional methods.
• Weeks 2 to 4—AI-assisted task completion: students attempted to complete the same composition task using the generative AI tool DALL-E.

The methodology ensured that participants completed both tasks in a controlled environment where their performance could be observed and evaluated systematically. This approach allowed for a meaningful comparison between traditional and AI-assisted composition techniques, providing valuable insights into how generative AI tools impact the creative process.

3.1. Week 1: Lecture and Traditional Composition Task

All 10 participants in the composition class were invited to attend a 30 min lecture on the fundamental principles of composition, including balance, contrast, unity, and emphasis. The lecture was followed by a brief discussion on how to apply these principles in a practical task. After, students were given a practical task (for detailed description, see Appendix A) that required them to create a basic composition following the principles they had learned. Students were given 1.5 h to complete the task (Figure 1).

3.2. Weeks 2–4: AI-Assisted Composition Task

The participants were introduced to the generative AI tool they would use for the second task. The computer with the ChatGPT premium version was provided to them. Those unfamiliar with the tools were given a brief demonstration of how to input prompts and adjust settings to generate visual content. The students were instructed to replicate the same composition task they performed in the first week, but this time using generative AI tools. They input prompts that described their intended compositions and attempted to refine the AI-generated images to match their original work (Figure 2).

Each participant was observed individually over the course of their 30 min task completion, during which they were encouraged to engage in a “think-aloud” process. This method allowed the researcher to document their thought processes, challenges, and task completion strategies in real-time. To systematically capture data, an observation protocol (Appendix B) was developed and utilized for each participant. The protocol was designed to record key parameters related to the students’ interaction with the AI tools and their approach to the task. Each protocol was filled in separately for individual participants to ensure detailed and personalized documentation.

3.3. Post-Task Questionnaire

After both the traditional and AI-assisted tasks were completed, each student filled out a questionnaire that measured their experiences. The survey comprised a total of 10 questions, structured to gather both quantitative and qualitative data regarding participants’ experiences with generative AI tools in the context of art education. The survey was divided into two types of questions: structured (multiple-choice items and Likert-scale statements ranging from “Strongly Disagree” to “Strongly Agree”) and open-ended. The structured questions focused on the following themes:

• Demographics and background: Age, gender, and prior experience with generative AI tools.
• AI tool usage: Frequency and purposes of using generative AI tools, such as ideation, concept development, and design refinement.
• Perceptions and attitudes: Participants’ views on the impact of AI tools on productivity, creativity, task efficiency, and design quality.
• Ease of use and accessibility: Assessing the usability, clarity, and learning curve of generative AI tools.
• Support and integration: Perceived support from peers, instructors, and the institution for using AI tools in art education.
• Future intentions: Plans to continue using AI tools in academic and professional work.

The survey included four open-ended questions to allow participants to elaborate on their experiences and provide nuanced feedback. These questions explored the following:

• The impact of AI tools on participants’ creativity.
• Challenges faced when integrating AI tools into their design workflows.
• Participants’ perceptions of AI’s potential role in their future careers as designers.
• Suggestions for improving the integration of generative AI tools in art and design education.

This mixed-methods approach ensured a comprehensive data collection process. The structured questions facilitated measurable comparisons and statistical analysis, while the open-ended responses provided rich qualitative insights into participants’ individual experiences and perspectives. The questionnaires were administered electronically through the Google Forms platform, ensuring ease of access for participants while maintaining a controlled and consistent method for data collection. Students were provided with the survey link in the first lecture. They were instructed to complete the questionnaire individually and were informed about the deadline to submit their responses within a specified timeframe. The questionnaire was designed to be anonymous.

While participants were assigned a respondent number at the start of the survey for organizational purposes, no personally identifiable information was collected, ensuring complete confidentiality. At the beginning of the questionnaire, a detailed informed consent section was included. This section clearly outlined the purpose of the research, the voluntary nature of participation, and the participant’s right to withdraw at any time without any negative consequences. The consent section explicitly stated the following: That the study aimed to investigate the role of generative AI tools in art education; that their responses would be used solely for academic research purposes; and that the data would be anonymized and securely stored to protect confidentiality. Participants were required to check a box to confirm their consent to participate before proceeding to the survey questions. This served as an acknowledgment that they had read and understood the research information and voluntarily agreed to participate. Students were given clear instructions on how to complete the questionnaire, including a brief explanation of each section and the approximate time required to complete it. They were also encouraged to provide thoughtful and honest responses, especially for the open-ended questions, to contribute valuable insights to the study.

The combination of surveys, structured observations, and participant feedback provided a comprehensive approach to understanding the integration of generative AI tools in art education. By utilizing a mixed-methods framework, the study ensured both quantitative and qualitative depth, capturing measurable data alongside rich, contextual insights. The observation protocol, think-aloud methodology, and survey responses ensured that participants’ experiences were documented systematically and ethically. These methods were designed to align with the research objectives, emphasizing the participants’ interactions with AI tools, their creative processes, and their perspectives on AI’s role in their education and future careers.

4. Results

This study collected data from 10 bachelor’s-level art program students regarding their experiences using generative AI tools, focusing on how these tools assist in the process of creating and learning about composition. The data were collected through practical tasks using both traditional observation and a questionnaire. The analysis focused on understanding the perceived efficacy of generative AI tools, the challenges faced, and the students’ overall attitudes toward their use in art education.

4.1. Observation

The observations of 10 student participants engaging with generative AI tools for artistic composition reveal distinct patterns in their interaction strategies, challenges, and emotional responses. The data indicate that students adopted iterative refinement techniques, faced difficulties in prompt formulation, and demonstrated varied levels of confidence in AI tool navigation.

All participants (10) engaged in an iterative process to refine AI-generated outputs, demonstrating a pattern of progressive prompt adjustment. The majority of students (8) initially provided broad, conceptual prompts, later refining them with additional specifications based on the output generated. For example, one student initially entered “create a geometric composition” but refined it to “place four red circles in the top-left corner with a white background”. This refinement process involved three to six iterations per participant, highlighting the trial-and-error nature of AI-assisted design.

Moreover, five students employed a layered approach, constructing compositions in stages (e.g., first generating a background, then adding objects, and finally adjusting details such as color and texture). However, the remaining students (5) relied on direct prompt modifications, without structuring their workflow into progressive steps. Across all participants, a common challenge was difficulty in translating artistic intent into precise text-based AI prompts. This issue was observed in multiple aspects. Students (7) struggled to specify precise placement of elements, resulting in unintended AI-generated compositions. One participant noted, “The AI isn’t fully capturing the symmetry I have in mind”. Six students reported that AI often added extra elements or misinterpreted requests for specific numbers of shapes. For instance, one participant explicitly requested six geometric shapes, but the AI generated additional unintended elements. Four participants experienced inconsistencies in maintaining a fixed color scheme across iterations, despite explicitly specifying color palettes.

Student confidence in navigating the AI tool was moderately high, with seven participants demonstrating competence in basic operations such as entering prompts and adjusting parameters. However, confidence declined when refining more complex compositions, particularly when precise alignment or proportional adjustments were needed. Three participants attempted to use seed values to maintain consistency across iterations yet noted that AI-generated outputs still exhibited randomness, making exact replication challenging. Five students displayed uncertainty when adjusting prompts, often relying on trial and error rather than structured refinement strategies. Eight students actively verbalized their thought process while working with the AI tool, making remarks such as “I need the shapes to feel balanced” or “This color doesn’t quite match what I intended”. However, four students expressed difficulty in translating these artistic observations into effective textual instructions.

The students exhibited neutral to mildly frustrated emotional responses throughout the design process. All participants (10) displayed engagement, but frustration was observed in specific instances. When AI-generated outputs deviated from expectations, six participants expressed visible signs of dissatisfaction (e.g., sighing, reattempting prompts multiple times, or verbally expressing disappointment). Seven participants reported a sense of accomplishment when AI-generated outputs progressively aligned with their vision. One participant stated, “It’s getting closer to what I want, but I wish I could tweak small details more easily”. Four participants adjusted their expectations after multiple iterations, accepting minor inaccuracies in shape placement or proportions as a trade-off for efficiency in generating compositions.

The degree to which the final AI-generated outputs aligned with students’ original artistic visions varied across participants. The majority (7) achieved only partial alignment, where the AI-generated composition captured core elements of their artistic goal (e.g., color scheme and general layout) but lacked precision in spatial relationships or object proportions. Two participants were unable to achieve their intended artistic outcome due to persistent discrepancies in AI-generated shapes and spacing, despite multiple refinements. Six students mentioned that AI-generated images could be exported and further refined manually in traditional design software, indicating that AI served as a starting point rather than a final design tool.

4.2. Questionnaire

The questionnaire responses were collected from the 10 students who participated in the study, offering insights into their familiarity with generative AI tools, the perceived impact of these tools on their design work, and their attitudes toward their use in the future. Participants ranged in age from 18 to 26, with the majority (6 students) in the 21–23 age group. The group consisted of six female and four male students.

Regarding familiarity with AI tools, seven students reported using AI primarily for idea generation and concept development, while fewer used them for tasks such as color scheme selection (four students) or finalizing designs (three students). Among the tools used, DALL-E was the most popular (four students), followed by Canva’s AI features (two students) and Stable Diffusion (one student). Notably, three participants indicated that they did not currently use generative AI tools, highlighting diverse levels of familiarity within the group.

The students agreed that generative AI tools enhanced their productivity, with an average rating of 4.3 out of 5, allowing them to complete design tasks more efficiently. Similarly, they rated AI tools highly in improving the quality of their work, particularly during the ideation phase, with an average rating of 4.5 out of 5. The majority (seven students) agreed that generative AI tools are most beneficial in the conceptual stage, as they provide inspiration and assistance in early design processes.

While eight students found AI tools user-friendly and expressed confidence in their ability to learn and operate them effectively, the average ease-of-use rating was lower, at 3.8 out of 5. This indicates that while students appreciated the tools’ potential, some faced a steep learning curve, which they overcame with practice. Challenges noted by participants included difficulty in aligning AI outputs with precise artistic instructions (three students), overwhelming variety of AI-generated outputs, making it hard to refine ideas (two students), technical issues with the tools’ functionality (two students). A few participants noted that mastering the use of generative AI tools had a steep learning curve, though this was overcome with practice. However, two students highlighted challenges, such as a steep learning curve and difficulty balancing AI-generated ideas with their own creativity, highlighting that while students appreciate the potential of generative AI, they find the learning curve somewhat challenging.

The students expressed positive attitudes toward the future role of AI in design. Seven out of ten students indicated their intent to use AI tools regularly in the future, with all participants acknowledging their usefulness in ideation and efficiency. Seven described AI as a “collaborator” rather than a replacement, emphasizing its potential to enhance creativity while underscoring the importance of maintaining human creativity and intuition. Despite optimism, some students expressed caution about over-reliance on AI tools, stressing the need to balance AI assistance with the cultivation of traditional artistic skills. Students provided actionable suggestions for better integrating AI tools into art education. Four students emphasized the need for hands-on workshops, while three recommended incorporating case studies of successful AI use in design. Additionally, two students suggested fostering collaborative projects that utilize generative AI tools to encourage teamwork and practical experience.

Table 1. Summary of Key Observed and Survey-Based Findings.

Finding	Key Observations and Survey Results	Implications for Art Education
AI for Brainstorming	Seven of ten students reported that generative AI was most useful for ideation rather than fine-grained composition details.	Suggests AI excels at sparking initial creativity but remains limited in executing nuanced artistic tasks.
Iterative Trial-and-Error	All 10 students iterated prompts three to six times to refine AI outputs.	Demonstrates that learning-by-doing (iterative refinement) is central to bridging the gap between AI output and artistic vision.
Prompt-Formulation Challenges	Five students struggled with specifying spatial relationships, proportions, and color consistency in their prompts.	Highlights the importance of prompt-engineering skills. Students need practice translating artistic intent into precise AI instructions.
Layered vs. Direct Workflows	Five participants used a layered approach (step-by-step building of an image), while five used direct prompts and immediate edits.	Reveals two distinct strategies for AI-assisted artmaking: structured layering vs. ad-hoc refinement, each with different learning curves.
Misalignment and Dissatisfaction	Four students expressed frustration when AI misinterpreted precise design instructions.	Indicates AI’s current limitations in fine control and the potential for user dissatisfaction when outputs deviate from intended results.
Basic Competence, Advanced Challenges	Seven students felt confident in basic AI operations but struggled with precise details or advanced adjustments.	Confirms that initial familiarity is achievable, but more specialized training is needed to handle complex design tasks effectively.
AI as Complementary, Not Replacement	Eight students described AI tools as enhancing creativity but not replacing human intuition or craftsmanship.	Suggests a positive view of AI as a “co-creator”, emphasizing that human ingenuity remains essential for artistic decision-making.
Future Intentions and Caution	Seven students intend to continue using AI, specifically for brainstorming but caution against over-reliance on AI-generated content.	Reflects students’ recognition of AI’s value in ideation while acknowledging the risk of eroding foundational skills if used too heavily.
Proposed Improvements	Four students requested hands-on workshops, three suggested case studies, and two advocated for collaborative projects.	Indicates a demand for structured support; educators could embed these methods into curricula to develop both AI and artistic competencies.

captures the key findings from the study, drawing on both observational data and questionnaire responses. The table is structured to align each key theme with supporting evidence, thereby illustrating how participants’ feedback, documented behaviors, and survey results interrelate.

5. Discussion

This study offers valuable insights into how art students utilize generative AI tools for artistic composition. While these tools support idea generation and concept development, challenges persist in achieving precision, control, and artistic intent.

One noteworthy factor behind these challenges is the difficulty of translating complex visual ideas into text-based instructions, which can be attributed both to a lack of prompt-engineering skills and to the limitations of language for describing relative or absolute positions, shapes, and sizes [1,5]. This twofold constraint often forced participants into iterative, trial-and-error workflows, as evidenced by their repeated refinement of prompts when AI outputs did not match their artistic visions. Our findings highlight the need for more structured guidance on prompt formulation in art education. This study addressed the following research questions:

1. 

What are the key challenges students face when using generative AI tools for artistic composition?

One major challenge was the difficulty in crafting precise prompts that accurately conveyed artistic vision to AI systems. Seven students struggled with spatial placement and proportions, frequently encountering unintended compositions due to AI misinterpretation. Three students attempted to use seed values to maintain consistency, but the randomness of AI-generated outputs made exact replication difficult. Students also found that AI often misinterpreted numerical requests (e.g., generating extra elements when specifying a fixed number of shapes). This highlights a critical gap in AI literacy, as effective prompt engineering plays a crucial role in enhancing the accuracy and relevance of AI-generated outputs. Research suggests that well-structured prompts can improve response accuracy rates from 85% to as high as 98% across various contexts, underscoring the significance of this skill in AI development and application [47]. However, novice users frequently rely on trial and error, which can lead to frustration and inefficiency. These findings emphasize the necessity of integrating AI literacy training into art education, equipping students with the ability to craft precise instructions that align with their creative intentions. It is worth noting that even experienced AI users may encounter hurdles when attempting to describe intricate compositions via text alone. For tasks requiring specific object placement and proportional relationships, language-based input can be inherently limiting, necessitating more robust interface design or multimodal feedback loops.

2. 

How do students refine and iterate AI-generated compositions to align with their artistic intent?

All participants engaged in iterative refinement, using multiple prompt adjustments to improve AI-generated results. Eight students initially provided broad conceptual prompts before progressively refining them with additional specifications. However, only half of the students structured their workflow systematically, while the others relied on intuitive, trial-and-error adjustments. Five students adopted a layered approach, generating elements in stages (e.g., background first, then objects, then details), while the remaining five relied on direct prompt modifications. Some students verbalized their thought processes while refining AI outputs, indicating an adaptive learning approach.

Many students found that mastering AI tools involved a steep learning curve, particularly for tasks requiring precise execution, such as intricate compositions. The complexity of AI-generated outputs sometimes led to inefficiencies in the refinement process, as students struggled to align results with their artistic vision. Some participants noted that AI-generated images often deviated from their intended outcomes, especially for nuanced or highly detailed compositions, highlighting the limitations of generative AI in fine-tuned artistic tasks. Such limitations may reflect the current stage of AI model development, where object relationships and spatial accuracy can be overlooked unless users input extremely detailed prompts or rely on advanced features [3].

Additionally, a few students expressed concerns about the potential drawbacks of overreliance on AI tools. While AI facilitated rapid ideation, some felt it reduced their engagement with foundational artistic principles, such as composition and spatial relationships. This suggests that while AI can enhance creative exploration, structured training in prompt formulation and systematic refinement strategies could help students better anticipate AI-generated outcomes and maintain a balance between automation and artistic intentionality [5]. Several studies corroborate the importance of balancing novel AI-driven practices with manual, hands-on design experiences, as students require fundamental skills and critical thinking to progress beyond automated outputs [4].

3. 

What are students’ attitudes toward the integration of generative AI tools in art education?

The majority (seven students) expressed a positive attitude toward AI’s role in art education, viewing it as a valuable collaborator rather than a replacement for traditional artistic techniques. All participants recognized AI’s usefulness in ideation and efficiency. On a scale of 1 to 5, most respondents rated AI tools highly (4 or 5) for enhancing productivity and improving design quality. This indicates that AI significantly streamlines the design process, particularly in visualization and experimentation. This finding aligns with Inie et al. (2023), who observed that a diverse range of creative professionals recognized generative AI’s potential to boost productivity, provide inspiration, and contribute to higher-quality artistic output [48].

Despite the perceived benefits, some students expressed concerns that excessive reliance on AI could hinder their development of foundational artistic skills. This concern aligns with Adeoye and Jimoh (2023), who emphasize that fostering creativity and innovation is essential for equipping learners with the skills needed to succeed in the 21st century [49]. Creativity is closely linked to problem-solving abilities and can be nurtured through opportunities for hands-on artistic exploration. However, the ease with which AI generates compositions may reduce student engagement with fundamental composition principles, highlighting the need for a balanced approach that integrates AI as a supportive tool rather than a substitute for traditional artistic techniques.

The students recommended more hands-on workshops and courses specifically focused on AI tools. Structured learning environments would reduce the learning curve and help students apply AI effectively. The respondents suggested integrating case studies and collaborative projects into art education to showcase successful uses of AI in art and design. This could deepen their understanding of AI’s potential for facilitating creative processes, especially in complex artistic tasks. Encouraging students to balance AI tools with traditional art techniques would enhance their understanding of composition. AI should serve as a complementary tool rather than a replacement for foundational skills.

These findings highlight the dual role of AI in art education: while it can foster ideation and broaden creative possibilities, it also introduces obstacles related to language-based inputs, model limitations, and potential overreliance. Future research could explore multi-modal AI interfaces, such as combining voice commands with sketches to mitigate the difficulties of translating intricate details or spatial relationships into text-based prompts. Incorporating real-time feedback loops in AI-based design platforms may further enable students to refine their compositions iteratively, thereby bridging the gap between verbal descriptions and artistic intent. Overall, the findings indicate that students perceive generative AI tools as useful aids in the ideation phase but face significant challenges in translating artistic intent into precise AI-generated compositions. While AI tools are seen as enhancing creativity and efficiency, their limitations in spatial control, color consistency, and detailed refinements create a sense of frustration. Most of the students (seven out of ten) view AI as a collaborator rather than a replacement for traditional artistic techniques, emphasizing the importance of maintaining human creativity and intuition in the design process.

6. Conclusions

This study examined the experiences of 10 bachelor’s-level art students using generative AI tool DALL-E for a composition task. Based on observations and questionnaire responses, the findings highlight both the advantages and limitations of generative AI tools in art education.

• The majority of students (seven) found generative AI useful for brainstorming and developing initial ideas, rather than executing detailed artistic compositions.
• All students (10) engaged in a trial-and-error process, progressively adjusting prompts to achieve desired results, with three to six iterations per composition.
• Five students struggled to translate their artistic intent into effective text prompts, particularly when specifying spatial relationships, proportions, and color consistency.
• Five students used a layered approach to build compositions in stages, while the other five modified direct prompts without structured workflows.
• Four students expressed dissatisfaction when the AI outputs did not align with their artistic vision, often due to the AI’s tendency to misinterpret precise design instructions.
• Most students (seven) demonstrated competence in basic AI operations; confidence declined when tasks required detailed adjustments or alignment.
• Eight students described AI tools as a complementary resource, enhancing creativity but not replacing human intuition and craftsmanship.
• The majority (seven) intend to use AI tools in the future, especially in the ideation phase, but caution against over-reliance on AI-generated content.
• The students suggested hands-on workshops (four students), case studies of AI in design (three students), and collaborative projects (two students) to enhance learning and application of AI tools.

These findings suggest that while generative AI has potential in art education, it requires structured integration with traditional artistic methods. While generative AI accelerates ideation and experimentation, educators and administrators must remain mindful of potential over-reliance on AI-driven outputs. Overuse of AI-generated content may reduce opportunities for students to develop hands-on expertise in foundational artistic skills such as color theory, perspective, and composition. Art programs should, therefore, balance AI-assisted tasks with traditional, manual projects that reinforce critical thinking and creative autonomy.

Skills development in prompt engineering and AI literacy will be increasingly essential for emerging art professionals. Students who can effectively translate artistic intent into clear, structured prompts and refine AI outputs are positioned to excel in an evolving creative landscape. However, integrating these skills into the curriculum requires structured guidance, hands-on workshops, and transparent rubrics that distinguish between legitimate AI collaboration and unethical shortcuts.

Ethically, the ease with which AI tools can generate finished pieces raises questions about academic honesty and attribution. As AI continues to expand its creative capacities, higher education institutions must implement clear guidelines that encourage honest reporting of both human and AI contributions.

With 10 participants drawn from a single course, the ability to generalize conclusions is limited. Future research should investigate how evolving AI capabilities can better align with artistic intent and explore larger participant samples over multiple semesters, comparing how different generative AI platforms (e.g., DALL-E, MidJourney, Stable Diffusion) are integrated into creative workflows.