Source-linked AI summary
From Style Replication to Style Exploration: Enabling Art Style Exploration with Analyze-Experiment-Resituate Framework
Wen-Fan Wang, TsaiHsuan Lin, Chi-Lan Yang, An-Ru Cheng, Bing-Yu Chen
TL;DR
Existing digital creativity tools offer limited support for exploring new styles beyond reproducing visual features. This paper proposes AER, an artist-guided framework for analyzing references, experimenting with stylistic possibilities, and resituating emerging styles; compared with direct style transfer, it increased artists’ perceived agency and reflection.
Problem
Digital creativity-support tools remain under-examined for style exploration and often reduce style to reproducible features optimized for fidelity rather than artistic practice.
Method
AER guides artists through analyzing technical and conceptual artwork elements, experimenting with controllable combinations, and reflecting through simulated social perspectives.
Results
AER produced significantly higher perceived agency, self-efficacy, and technology-supported reflection than direct style transfer among 16 artists.
Takeaways & Limitations
The findings support designing GenAI style-exploration workflows around artists’ analysis, experimentation, reflection, and creative agency rather than output reproduction alone.
Takeaways & Limitations
AER raises ethical concerns around style mimicry, copyright, consent, attribution, and credit when referencing other artists’ works.
Abstract
from arXiv · showhide
Art style is a signature of professional digital artists that develops through repeated experimentation, reflection, and adaptation. While generative AI (GenAI) can reproduce styles with high fidelity, current tools provide limited support for exploring new stylistic directions and may encourage style replication over exploration. To address this gap, we propose Analyze-Experiment-Resituate (AER), a framework for AI-assisted style exploration derived from interviews with 10 professional digital artists. Rather than prioritizing visually appealing outputs alone, AER supports three core practices of style exploration, including interpreting references, trying out stylistic possibilities, and reflecting on how emerging styles may be received. Specifically, AER enabled artists to (1) analyze artworks into interpretable stylistic elements, (2) have controllable experimentation guided by their own choices, and (3) resituate emerging styles through simulated social perspectives. We implemented AER in a prototype system and evaluated it in a controlled study with 16 artists. Compared with a direct style-transfer workflow, AER increased artists' agency and reflection as they pursued new stylistic directions. A two-week field study with four artists revealed how the AER framework influenced daily style exploration, such as reflection, experimentation, and stylistic decision-making at each stage. We discuss opportunities and challenges in designing AI-assisted style-exploration workflows, and outline implications for future artistic support tools.
1 Introduction
The paper frames style exploration as an iterative, reflective practice of studying references, experimenting with techniques, and adapting artistic choices, rather than merely reproducing visual features. It investigates how GenAI can support artists’ interpretation, decision-making, and creative agency during this process.
- Motivation: Art style expresses artistic identity, emotional intent, and creative thinking through recurring visual and conceptual choices.For professional digital artists, style is also central to long-term creative development.
- Motivation: Digital tools and online platforms accelerate style exploration by expanding access to references and enabling faster experimentation.GenAI further supports rapid ideation and image refinement.
- Problem: AI research often treats style as aesthetic features to copy or transfer, whereas artists experience it as evolving, situated, and requiring reflection and iterative exploration.Current GenAI tools tend to prioritize stylistic outputs over artists’ interpretation and decision-making.
- Research approach: The work investigates GenAI-supported style exploration through three studies involving professional illustrators and concept artists.The studies examine artists’ existing practices, the AER framework versus direct AI-generated style transfer, and its use in practice.
- Contributions: The contributions include an empirical account of artists’ style-exploration practices, an interaction framework for AI-assisted exploration, and a demonstration of supporting analysis, experimentation, and reflection without compromising creative agency.The framework is informed by the formative study and examined through controlled and field studies.
2 Related Work
Prior work commonly treats style as transferable visual features and emphasizes artifact production, fidelity, or specific creative tasks. This leaves process-centered, long-term style exploration and artist agency comparatively under-supported, motivating the proposed framework.
- Style exploration: Artists explore style through a continuous, recurring, non-linear process of analyzing prior work, making art, adapting style, and reflecting on feedback.This process can span years and involves dialogue among reference works, tools, and creative media.
- Style exploration: Digital-art research has primarily focused on producing specific artifacts, leaving digital support for exploring new styles under-examined.The paper shifts creativity-support tools from artifact production toward the process of style evolution.
- Style transfer: Computational style-transfer research models style as reproducible visual attributes such as color and texture, prioritizing transferable features and output fidelity.This tradition includes extracting stylistic components from references and applying them to other images.
- Limitations of existing tools: Output-centric GenAI tools can obscure artists’ intentions, reduce agency, override original expression, and standardize outputs toward stylistic homogenization.Opaque intermediate representations can also make it harder for artists to understand why specific outputs are produced.
- Research gap: Existing HCI systems emphasize ideation, generation, fine-grained control, or artifact-level reflection rather than longer-term exploration of new styles through simulated social perspectives.The proposed work therefore uses a process-centered framework to scaffold stylistic exploration with GenAI while maintaining creative agency.
3 Method
The research used a mixed-methods design spanning formative, controlled experiment, and field studies to examine artists’ style exploration and AI-assisted workflows. Interview data underwent iterative thematic analysis, and all studies received institutional ethical approval.
- Study Design: Mixed methods combined semi-structured interviews, questionnaires, and log data to examine artists’ practices, GenAI perspectives, and workflow influences.
- Ethics: All studies were approved by the authors’ institutional ethical review board.
- Study Design: The research comprised three studies: formative (RQ1), controlled experiment (RQ2), and field study (RQ3).
- Data Analysis: Two authors initially coded transcribed interviews, then the research team iteratively refined themes until consensus using thematic analysis.
4 Formative Study: Artists’ Style Development: Practices, Challenges, and Design Goals
Interviews with 10 professional digital artists characterized style development as a time-intensive cycle of reference analysis, experimentation, adaptation, and socially informed reflection. They also revealed that existing GenAI tools obscure artistic rationale and constrain agency, motivating three design goals for AI-assisted style exploration.
- 4.1 Participants: 10 professional digital artists participated, including four concept artists and six illustrators with a mean of nine years’ experience.All participants were based in East Asia; ages ranged from 23–38, with six male and four female artists.
- 4.2 Study Procedure: Interviews lasted 1–1.5 hours via Google Meet and focused on participants’ personal work, project experiences, and style-development journeys.Participants prepared representative artworks and experiences beforehand, and consent was obtained before interviews.
- 4.3.1 Practices and Challenges in Style Development: Artists’ style exploration began with personal preferences, reference collection, and attempts to imitate admired works.They encountered references through active searching or passive browsing, using admired styles as sources for learning and experimentation.
- 4.3.1 Practices and Challenges in Style Development: Artists analyzed references by decomposing composition, linework, color, lighting, rendering style, and shape language to understand both how and why styles work.They then experimented, adapted techniques, and combined them with personal preferences and habits to develop their own styles.
- 4.3.1 Practices and Challenges in Style Development: Feedback from clients, viewers, and peers influenced stylistic decisions, while unmet expectations and weak resonance prompted reflection on personal style.Social responses could steer artists toward emphasized stylistic features or motivate reconsideration of their direction.
- 4.3.2 Challenges in Using GenAI in Style Development: Artists reported that GenAI outputs were visually appealing but difficult to learn from because they concealed technical processes, artistic intent, and emotional meaning.None of the 10 participants used GenAI or style transfer for style exploration, and AI references were seen as remixing unattributed sources without clear rationale.
- 4.4 Design Goals: GenAI’s complete designs reduced opportunities for intervention and contemplation, undermining artists’ creative agency and independent thinking.These limitations motivated goals to preserve agency, make stylistic reasoning explicit, and integrate social perspectives into exploration.
5 Analyze, Experiment, and Resituate: A Framework for Creative-support Tools
The AER framework structures AI-assisted style exploration into Analyze, Experiment, and Resituate stages, integrating interpretable reference analysis, artist-guided variation, and simulated social feedback. Its proof-of-concept system operationalizes this workflow for empirical evaluation.
- System implementation: The proof-of-concept system operationalizes AER as an AI-assisted style exploration workflow and enables empirical evaluation through user studies.The system uses three interface panels corresponding to Analyze, Experiment, and Resituate.
- Analyze: Analyze decomposes reference artworks into technical and conceptual stylistic elements for artists to interpret.Technical elements include composition, shape language, linework, lighting and atmosphere, color palette, and rendering style; conceptual elements include emotional expression, cognitive process, and intended messages.
- Experiment: Experiment lets artists choose technical or conceptual elements and generate interpretable stylistic variations with explanations of each change and rationale.This balances active control with diverse outputs while helping artists understand the underlying stylistic logic.
- Resituate: Resituate simulates feedback from valued social roles, including professional artists, trending audiences, and fans, to support stylistic reflection.The simulated professional artist provides constructive critique, the trending audience addresses immediate social-media appeal, and the fan addresses stylistic consistency and recognition.
- Framework overview: Together, the three stages integrate reference decomposition, active variation, and social feedback into professional artists’ style-exploration workflow.The framework is designed for artists with an established style to explore new styles by integrating their existing style with others they wish to pursue.
6 Controlled Experiment Study: Exploring and Evaluating AER Framework … 6.4 Measurement
A within-subjects study with 16 professional digital artists compared an AER-embedded system with direct style transfer to examine agency and reflection during style exploration. The study combined controlled style-exploration tasks, validated questionnaires, and semi-structured interviews.
- 6 Controlled Experiment Study: Exploring and Evaluating AER Framework: 16 professional digital artists participated in a within-subjects study examining how AER influenced agency and reflection during style exploration.The study addressed research question RQ2.
- 6.1 Experiment Design: The study compared the AER-embedded system with a direct style transfer baseline designed to reflect common AI-assisted style-copying workflows.The baseline used a similar interface and was selected because direct style transfer is widely used in research and practice.
- 6.1 Experiment Design: The baseline allowed participants to upload style and content references, enter prompts, and view generated results using the same Flux-Kontext-Pro model as AER.It also used the same prompt structure as the AER-embedded system.
- 6.1 Experiment Design: Participants explored styles using their own current-style works or projects and admired reference artworks they wished to investigate stylistically.The resulting outputs served as inspirational references rather than final artworks.
- 6.1 Experiment Design: The order of conditions was counterbalanced, and participants were encouraged to think aloud while exploring potential stylistic directions.Possible outcomes included AI-simulated illustrations, insights, and next steps for style development.
- 6.2 Participants: Participants were 16 professional digital artists aged 23–38, including 6 concept artists and 10 illustrators with 4–15 years of experience.Their mean experience was 7.25 years, and each received 35 $USD for the 2-hour study.
- 6.3 Procedure: Each condition included a 10-minute tutorial, 30–40 minutes of style exploration, and a 5-minute post-task questionnaire, followed by a 30-minute interview.The study began with a 10-minute briefing and concluded with a semi-structured interview.
- 6.4 Measurement: Participants completed Agency, Creative Self-Efficacy, and TSRI questionnaires using 7-point Likert scales, supplemented by interviews on agency, creative process, future style direction, and AI-assisted approaches.Scale reliability was high: Agency α=.86, Creative Self-Efficacy α=.90, TSRI α=.91, with Insight α=.86, Exploration α=.82, and Comparison α=.78; analyses used paired t-tests and Cohen’s d after Shapiro-Wilk normality checks.
6.5 Findings
Compared with direct style transfer, AER increased artists’ perceived agency, self-efficacy, and technology-supported reflection. Artists attributed these benefits to traceable and controllable choices, interpretive experimentation, social feedback, and alignment with everyday creative practices, while also calling for workflow flexibility.
- Agency: AER significantly increased perceived agency compared with direct style transfer (t[15] = 2.95, p= .010, d= 0.85, mean difference = 0.88).Artists linked this to traceable decisions, controllability, and alignment with their artistic identity.
- Reflection and exploration: AER significantly increased self-efficacy (t[15] = 2.90, p= .011, d= 0.88, mean difference = 1.00) and technology-supported reflection (t[15] = 3.21, p= .006, d= 0.92, mean difference = 0.96).The TSRI results also showed stronger ability to gather insight, while Analyze helped artists articulate stylistic elements and identify new directions.
- Experimentation: Artists used generated images as practical references for clarifying stylistic direction and anticipating challenges during experimentation.They reported that this could prevent wasted time pursuing unproductive stylistic paths.
- Resituate: Resituate feedback helped artists identify distinctive traits that might be lost when shifting styles and reintroduce personal characteristics into emerging work.Professional artist feedback provided constructive insights, while fan feedback highlighted what made artists’ work distinctive.
- Workflow fit and flexibility: Artists found AER aligned with daily practices across ideation, client pitching, post-project exploration, and teaching, but requested flexibility to skip stages.Some artists found Analyze sufficient for inspiration, while others skipped Resituate until they had created work themselves.
7 Field Study
A two-week field study with four professional digital artists examined how AER fit into ongoing style exploration. Participants integrated the stages nonlinearly, using Analyze, Experiment, and Resituate to support reflection, experimentation, and stylistic decision-making with varying usefulness.
- Study design: Four professional artists participated: two environment concept artists and two illustrators, each receiving approximately $220 USD.Participants had 4–10 years of experience and ranged from 24 to 31 years old.
- Study design: The field study deployed AER as a web-based system for two weeks, encouraging approximately one-hour daily sessions with flexible make-up usage.Participants completed daily diary entries and end-of-study semi-structured interviews grounded in those diaries.
- Usage patterns: Mean total usage was 9.63 hours, with participants spending 64.78 % in Experiment, 25.85 % in Analyze, and 9.37 % in Resituate.Artists did not use the stages in a strictly linear manner.
- Analyze: Analyze externalized tacit stylistic judgments by helping artists revisit principles, notice overlooked details, and clarify what they valued in references.Some participants found its labels insufficient because they identified visible elements without explaining their contextual effects on balance and emphasis.
- Experiment: Experiment made alternative stylistic directions concrete and comparable, helping artists identify missing elements, understand failed styles, and recognize interactions among stylistic choices.Participants used generated images as reusable repositories of stylistic elements, solutions, and cues rather than committing to a single output.
- Resituate: Resituate prompted critical comparison for some artists but was used less often because simulated perspectives lacked clear reference value and assessed rough outputs as completed works.Participants increasingly used AER to reinterpret references structurally and combine references for reusable details, rather than copying surface appearance.
8 Discussion
The discussion positions AER as an interpretable, artist-driven workflow that shifts GenAI-assisted practice from style replication toward reflective style exploration. It also identifies risks involving automation, contextual misunderstanding, feedback transparency, and ethical concerns around style mimicry and attribution.
- Interpretability and control: AER embeds interpretability and control into GenAI-assisted style exploration, addressing artists’ concerns about unpredictable image-generation processes.Participants described crafting prompts and then waiting for whatever GenAI would show them, whereas AER incorporates interpretive support and user control.
- Reflective creativity: Artists engaged in reflection-in-action with AER, continuously shaping exploration and shifting their mindset from style replication toward style exploration.Analyze infers stylistic intent, while Resituate provides interpretive refraction that encourages questioning assumptions, justifying decisions, and reflecting on stylistic goals.
- Artist agency: Artists flexibly adapted AER’s stages to their practices, positioning themselves as the primary drivers of the creative process.Participants did not treat the workflow as strictly linear and identified applying Analyze to prior artworks as a potential source of inspiration.
- Risks and future directions: The discussion identifies risks from AI-assisted style exploration, including offloading analytical skills, context-insensitive interpretations, opaque feedback, style mimicry, copyright risks, and misattribution.Proposed directions include preserving interpretive reasoning, incorporating cultural and historical context, explaining feedback mechanisms, and embedding attribution, consent, and credit.
- Broader applicability: AER’s three-part pattern—studying references, experimenting, and seeking feedback—could extend beyond visual art to music, writing, product design, and other creative fields.The discussion frames this pattern as a process aligned with creative practice rather than reasoning collapsed into an end-to-end system.
A Appendix A: Participants Demographic
Appendix A summarizes participant demographics, including age, identity, GenAI tools, and study participation.
- Table 1 reports participants’ age, identity, GenAI tools, and study participation.
B Appendix B: Questionnaire for Controlled Experiment Study
Appendix B presents the questionnaire used in the controlled experiment study, measuring agency, self-efficacy, reflection, exploration, and comparison with other artists. The measures assess participants’ perceived control, confidence, insight, reflection, enjoyment, progress review, and social comparison during style exploration.
- Agency: The Style Development Agency Scale measures perceived control, authorship, intentional guidance, decision-making, and direction of the style exploration process.Items ask whether participants control style decisions, author creative choices, pursue directions aligned with their intentions, and plan the process from beginning to end.
- Self-Efficacy: Style Exploration Self-Efficacy measures confidence in achieving personal exploration goals, handling difficult tasks, obtaining important insights, and pursuing chosen style directions.The items assess perceived capability during system use, including confidence in accomplishing difficult exploration tasks and exploring any intended style direction.
- Reflection and Exploration: TSRI measures whether the approach prompts changes in style exploration, new strategies for overcoming development challenges, enjoyment, and understanding of current and past progress.The questionnaire includes reflection items about changing one’s approach and gaining ideas, alongside exploration items about enjoyment, overview, and reviewing past experiments.
- Comparison: The Comparison scale measures reflection on style exploration with other artists, discussion of the process, and consideration of how one’s exploration compares with others’.These items focus on socially situated reflection about exploration results and processes.
C Appendix C: Figures
Appendix C illustrates the two workflows: direct style transfer combines a user’s artwork with a reference style through selectable style-tweaking outputs, while AER analyzes references, generates variations and explanations, and simulates social feedback.
- Direct Style Transfer: The direct style-transfer interface accepts a personal artwork, reference artwork, and prompt, then displays four selectable style-tweaking artworks.The outputs merge the user’s artwork content with the reference artwork’s style.
- AER Workflow: AER analyzes referenced artworks into conceptual and technical aspects, generates image variations with explanations, and simulates three social roles to provide feedback.Participants used this workflow to explore new styles and select image variations for feedback.