PaperlayerView exact revision on arXiv

Source-linked AI summary

Drawing the Line: Where AI Guidance and Human Creativity Meet in Emotion-Driven Comic Storyboarding

Jocelyn Shen, Isabella Pu, Alessandro Briseño, Hyun Kim, Fiona Lu, Sharifa Alghowinem, Cynthia Breazeal, Hae Won Park

arXiv:2610.09155v1cs.HC

TL;DR

Generative models can produce visually faithful artwork but may not convey the emotional authenticity important to visual storytelling. EmoToon uses intended feelings to guide comic storyboard generation and iteration. In a study, AI assistance improved perceived expression and aesthetics but reduced perceived creativity and ownership.

  • Problem

    Generative models may produce technically strong images but struggle to convey lived emotional experience, a gap that matters in emotionally driven visual storytelling.

    Source: §I · p. 1

  • Method

    EmoToon is a storyboard probe that uses character descriptions, panel scripts, and intended feelings and takeaways to generate panels that users can revisit and iterate on.

    Source: §III · p. 3, B. EmoToon Design · p. 4

  • Results

    EmoToon improved perceived emotional expressiveness and aesthetic quality relative to freehand drawing, but participants reported lower creativity and ownership.

    Source: §VI · p. 7

  • Takeaways & Limitations

    Emotion- and composition-rich references can help novices externalize affect and produce more polished storyboards, while high-level structural proposals may reduce perceived agency.

    Source: §VI · p. 7

    The study’s scope was non-expert creators using EmoToon for comic storyboarding, and the reported benefits came with lower perceived creativity and ownership.

    Source: §VI · p. 7, §VII · p. 7

Abstract

from arXiv · show

While generative AI models can produce visually faithful artwork, they often fall short in conveying emotional authenticity--a key driver of human expression. In visual storytelling, particularly comic storyboarding, this gap becomes pronounced: effective storyboards require both technical knowledge (e.g., anatomical accuracy or scenic composition) and emotional insight (from lived experience). We explore how human-AI collaboration can support emotion-driven creativity, where the user's feelings guide generation and emotional resonance is the goal. We present EmoToon, a technology probe that helps non-professional artists generate sketch-like storyboards and iterate on visual ideas. In a controlled study with N=25 participants, we find that AI assistance significantly improves emotional expression, aesthetic quality, and exploration, but reduces users' creative ownership. Our findings offer broader insights for human-AI co-creation in the domain of comic storytelling, emphasizing balance between output quality and user freedom, and raising new challenges for image generation models in this domain.

I. INTRODUCTION

The paper examines how AI can provide technical guidance for emotionally authentic comic storyboarding, particularly for non-professional creators. It introduces EmoToon and studies how AI assistance affects storyboard creation.

I. INTRODUCTION

  • AI can support technical tasks such as anatomical accuracy, but models cannot replicate the emotional richness of people’s lived experiences.Prior studies also find that AI-generated art evokes less emotion and narrative than human-made art.Source: §I · p. 1
  • The paper frames emotion-driven creativity as user feelings guiding generation, with emotional resonance as a goal of the output.It investigates balancing technical guidance for non-professional artists with preserving human emotional authenticity.Source: §I · p. 1
  • Non-professional creators may have emotionally resonant stories but lack the technical vocabulary, execution skills, or training to turn them into visual assets.The authors focus on this group to investigate making emotionally driven visual storytelling accessible to everyday creators.Source: §I · p. 1
  • EmoToon is a technology probe that turns storytellers’ intended emotions into drawing inspirations for comic storyboards and supports revisiting ideas during drawing.The study assessed self-perceived emotional expression, aesthetic quality, creative freedom, and ease of learning against freehand drawing with 25 participants.Source: §I · pp. 1–2, Figure 1 · p. 2

II. RELATED WORK

Prior work documents skepticism toward AI art alongside growing interest in AI as a creative assistant. Co-creative systems span multiple domains, but emotionally expressive visual storytelling for non-experts remains underexplored.

A. Human Perceptions on Creative AI

  • Artists and the public question AI art’s authenticity and creativity, and viewers often rate AI-labeled work lower than human-made work.Some artists nevertheless consider AI for creative assistance, especially supporting tasks rather than core creative decisions.Source: A. Human Perceptions on Creative AI · p. 2

B. Human-AI Co-Creation Interfaces

  • Creative AI systems support work in domains including music, data science, calligraphy, engineering, animation, design ideation, and writing.Some act as collaborators that contribute sketches or encourage iteration, while others scaffold technical skills or automate repetitive work.Source: B. Human-AI Co-Creation Interfaces · p. 2
  • Storytelling systems visualize emotional trajectories or transform stories into scenes, but emotionally expressive visual storytelling for non-experts remains little studied.Prior work includes Talebrush, StoryDrawer, and end-to-end pipelines for visual stories.Source: B. Human-AI Co-Creation Interfaces · pp. 2–3

C. Technology for Supporting Emotional Visual Expression

Existing computational emotion tools often do not address narrative visual techniques central to comics, and a notable sketching tool focuses on faces. EmoToon probes AI-assisted comic storyboarding with emotional intent as an input.

C. Technology for Supporting Emotional Visual Expression

  • Computational emotion-to-visual systems often focus on abstract representations rather than narrative techniques such as gesture, framing, and scene composition.These techniques are central to emotional storytelling in comics.Source: C. Technology for Supporting Emotional… · p. 3
  • EmoG generates emotional facial expressions for a drawn character but is limited to individual faces and six basic emotions, without narrative context or broader visual strategies.Those strategies include body language, staging, and composition.Source: C. Technology for Supporting Emotional… · p. 3
  • EmoToon treats user affect as a generative input and probes AI-assisted comic storyboarding for conveying complex emotions in figurative form.Its suggestions help non-experts translate emotional narrative intent into visual panels, and the study evaluates expression, aesthetics, creativity and ownership, and ease of learning.Source: C. Technology for Supporting Emotional… · p. 3

III. EMOTOON TECHNOLOGY PROBE DESIGN

  • The web-based probe takes character descriptions, a panel-divided comic script, and an intended feeling and takeaway, then generates one storyboard panel per input panel.It was developed to investigate how AI assistance affects emotional expression in co-created comic storyboards.Source: §III · p. 3

A. Visual Design Elements for Emotion Expression in Storyboarding

The design draws on composition and character expression as visual elements tied to emotion. EmoToon makes intended feeling an input, supports panel regeneration, and uses a staged image-generation pipeline.

A. Visual Design Elements for Emotion Expression in Storyboarding

  • The authors identify composition and facial or gestural character expression as storyboard elements tied to emotional expression.Effective composition is defined as alignment between panel arrangement and conveyed emotion; effective character expression concerns how clearly faces or gestures communicate characters’ emotional states and intentions.Source: A. Visual Design Elements for Emotion… · pp. 3–4

B. EmoToon Design

  • EmoToon asks users to state the intended feeling and takeaway, anchoring generation in expressive intent before detailed storyboards are drawn.The design also provides panel regeneration to support experimentation and revision.Source: B. EmoToon Design · p. 4
  • GPT-4 turns the script, character descriptions, and intended emotion into a detailed DALL-E prompt specifying camera angle, composition, expressions, and poses.DALL-E generates a manga-style image for each panel, which an Image-to-Line-Drawings model converts into a storyboard panel.Source: B. EmoToon Design · p. 4
  • The interface collects character descriptions, a panel-divided comic script, and intended feeling and takeaway, then presents generated panels with script editing and regeneration options.The pipeline is illustrated with example outputs.Source: B. EmoToon Design · p. 4, Figure 3 · p. 4

IV. USER STUDY

The study compared EMOTOON-guided storyboarding with freehand drawing among 25 non-professional artists. Participants created separate personal-story storyboards in both conditions, with story-condition pairings and condition order counterbalanced.

IV. USER STUDY

  • The study compared EMOTOON-guided storyboard drawing with a FREEHAND baseline among 25 participants.Participants had an interest in art, varied in drawing and storytelling experience, and none were professional artists.Source: §IV · p. 4, A. Participants · p. 4

B. Study Design and Procedure

  • Participants wrote two personal stories, each with a desired feeling and takeaway, and made a four-panel storyboard for an important life moment.They completed this pre-intervention task after a researcher introduced storyboarding and demonstrated the tool.Source: B. Study Design and Procedure · p. 4
  • Each participant used separate narratives for EMOTOON and FREEHAND; story-condition pairings and condition order were counterbalanced and randomized.The FREEHAND comparison isolated the probe's effect; reference searching and other complex workflows were excluded to prioritize internal validity over ecological realism.Source: B. Study Design and Procedure · p. 5
  • After each task, participants completed a survey and later took part in a 15-minute interview; they produced 50 storyboards in total.Each participant created two storyboards.Source: B. Study Design and Procedure · p. 5

C. Hypotheses

The study tested four expectations about EMOTOON versus freehand drawing, covering emotional expression, aesthetics, creative freedom, and learning. Surveys, interviews, interaction data, and artifacts were used to assess these dimensions.

C. Hypotheses

  • The hypotheses predicted better emotional expression, aesthetic satisfaction, and learning with EMOTOON, but no perceived difference in creative freedom.The four hypotheses also included whether users felt EMOTOON better conveyed the story's feeling than FREEHAND.Source: C. Hypotheses · p. 5

D. Data Collection and Analysis

  • Participants rated storyboard expressiveness, aesthetic appeal, creativity, and learning-curve measures on a 1–7 Likert scale.The measures included ownership and degree of freedom, as well as ease of exploration, confidence, composition, and character expressions.Source: D. Data Collection and Analysis · p. 5
  • Researchers recorded and transcribed post-condition interviews, then had two raters independently code interview utterances and artifacts.Interview coding reached Cohen’s κ = 0.67, and artifact coding fell in the stated substantial-agreement range of 0.61 < κ < 0.80.Source: D. Data Collection and Analysis · p. 5
  • Fig. 4 shows example participant-drawn panels across EMOTOON and FREEHAND conditions.The figure provides examples of participant artifacts rather than quantitative results.Source: Figure 4 · p. 5, D. Data Collection and Analysis · p. 5
  • Fig. 5 presents quantitative metrics for self-perceived emotional expression, aesthetic quality, creativity, and learning curve across EMOTOON and FREEHAND.Yellow highlighting marks significant differences.Source: Figure 5 · p. 5

V. RESULTS

EMOTOON improved participants’ self-rated emotional expressiveness compared with FREEHAND, while interviews and artifacts show how generated visual references informed their panels. Participants also reported limits in emotional nuance and progression across panels.

1) Quantitative Insights:

  • EMOTOON storyboards scored higher for EMOTIONALLY EXPRESSIVE than FREEHAND drawings (M = 5.2 vs M = 4.12; p = 0.01), but not for ART AND STORY MATCH.The comparison involved N = 25 participants; the expressiveness result had W = 27 and r = 0.50.Source: 1) Quantitative Insights · p. 5

2) Qualitative Insights:

  • Generated references for facial expressions, gestures, and panel composition helped participants visualize emotional states they struggled to depict.Participants described the tool as helping them show emotions they found hard to imagine drawing themselves.Source: 2) Qualitative Insights · pp. 5–6
  • In 23 of 25 storyboards, at least two panels drew emotional inspiration from generated examples, including facial expressions, body language, or composition.Source: 2) Qualitative Insights · p. 6
  • Participants described limits in emotional nuance and flow, including exaggerated styles that missed intended tone and inconsistent emotional progression across panels.Source: 2) Qualitative Insights · p. 6

B. H2: Effects on Aesthetic Quality

EMOTOON improved ratings of storyboard aesthetics and idea exploration, but did not yield significantly more creative final storyboards. Participants also reported feeling less creative, with lower ownership and freedom, while using EMOTOON.

B. H2: Effects on Aesthetic Quality

  • AESTHETIC ratings were higher with EMOTOON (M = 4.76) than FREEHAND (M = 3.2; p = 0.0007).The comparison had W = 15, N = 25, and r = 0.68.Source: 1) Quantitative Insights · p. 6
  • HELPFUL FOR EXPLORATION ratings were higher with EMOTOON (M = 4.92) than FREEHAND (M = 3.8; p = 0.04), while EASE OF EXPLORATION did not differ significantly.For EASE OF EXPLORATION, EMOTOON scored M = 4.68 and FREEHAND M = 4.08.Source: 1) Quantitative Insights · p. 6
  • Final storyboards were not significantly more creative with EMOTOON (M = 4.76) than FREEHAND (M = 5.44), and participants felt less creative while drawing with EMOTOON (M = 5.04 vs M = 5.76).The felt-creativity difference had W = 18.5, p = 0.008, N = 25, and r = 0.53.Source: 1) Quantitative Insights · p. 6
  • Participants reported lower OWNERSHIP (M = 4.88 vs M = 5.92) and DEGREE OF FREEDOM (M = 5.2 vs M = 5.8) with EMOTOON than FREEHAND.OWNERSHIP had W = 14, p = 0.008, N = 25, r = 0.53; DEGREE OF FREEDOM had W = 42, p = 0.05, N = 25, r = 0.39.Source: 1) Quantitative Insights · p. 6

2) Regenerate Feature:

EMOTOON eased storyboarding and supported exploration for some participants, while use of regeneration varied and AI-generated suggestions sometimes constrained creative ownership. Participants also reported greater ease of use and confidence than with freehand drawing.

2) Regenerate Feature:

  • Regeneration use varied: some participants repeatedly generated alternatives to explore ideas, while others preferred a single generated reference.Participants described iteration as a way to find an idea or option they liked more.Source: 2) Regenerate Feature · p. 6, Figure 6 · p. 6

3) Qualitative Insights:

  • Generated examples supported brainstorming and exploration, but some participants felt anchored by initial outputs and reported reduced ownership; others adapted or rejected suggestions.Source: 3) Qualitative Insights · p. 6

1) Quantitative Insights:

  • EMOTOON received higher LEARNING scores (M = 5.2, SD = 1.41) than FREEHAND (M = 4.2, SD = 1.63), but the difference was not significant.Source: 1) Quantitative Insights · pp. 6–7
  • EMOTOON was significantly easier to use (M = 5.44, SD = 1.33) than FREEHAND drawing (M = 4.64, SD = 1.52), and users felt significantly more confident using EMOTOON (M = 5.56, SD = 1.36) than FREEHAND drawing (M = 4.28, SD = 1.70).Ease of use: W = 33.5, p = 0.02, N = 25, r = 0.46; confidence: W = 18.0, p = 0.005, N = 25, r = 0.56.Source: 1) Quantitative Insights · p. 7

2) Qualitative Insights:

Participants described EMOTOON as lowering technical barriers, while the broader findings point to a trade-off between expressive, polished storyboards and users’ perceived agency. The discussion identifies sequential emotion and authorship as challenges for future co-creative tools.

2) Qualitative Insights:

  • Nineteen of 25 participants said EMOTOON simplified storyboarding, especially technically difficult poses, perspectives, and consistent environments across panels.Participants also linked visual references to greater confidence and experimentation with styles and compositions.Source: 2) Qualitative Insights · p. 7

VI. BROADER DISCUSSION, LIMITATIONS, AND FUTURE DIRECTIONS

  • EMOTOON improved perceived emotional expressiveness and aesthetic quality, ease of use, and confidence relative to FREEHAND, but reduced perceived creativity and ownership.The authors describe emotion- and composition-rich references as helping novices externalize affect and produce more polished storyboards, while structural proposals may anchor authorship and reduce agency.Source: §VI · p. 7
  • Conditioning generation on a “feeling and takeaway” helped participants translate emotional intent into facial expression, gesture, and composition choices.The authors argue that storyboard emotion unfolds across framing and panel progression, not only within a face.Source: §VI · p. 7
  • Participants reported that per-panel emotional assistance could disrupt coherence across panels through abrupt tone shifts or melodramatic spikes.The discussion proposes sequential reasoning that couples text, layout, and affect across panels.Source: §VI · p. 7
  • The discussion proposes emotional-arc visualizations, compositional alternatives, and composition palettes to help users explore framing, panel rhythm, and spatial weight.Source: §VI · p. 7
  • Even sketch-like AI references were associated with lower ownership and freedom; high-level choices such as composition and pose may shape authorship perceptions as much as surface polish.The authors suggest comparing AI references with conventional references and developing ownership measures for visual, spatial creation.Source: §VI · p. 7
  • Participants questioned whether increased confidence reflected actual learning or outsourcing; the authors suggest guided and minimal-support modes for different users.Suggested guided modes include explanatory overlays and prompts to modify references; minimal modes could restrict AI to thumbnails, composition alternatives, or mood boards.Source: §VI · p. 7

VII. CONCLUSION

The authors conclude that EMOTOON supports emotional expression, aesthetic quality, exploration, and confidence, while AI assistance can reduce ownership and freedom. They also identify limits to generalizability and unresolved ethical concerns around generative media.

VII. CONCLUSION

  • In a study with 25 users, EMOTOON aided self-perceived emotional expression, aesthetic quality, creativity exploration, and confidence compared with FREEHAND, but participants reported significantly lower ownership and freedom.The authors suggest varying the abstraction of artistic support to increase users’ sense of creative voice.Source: §VII · p. 7

ETHICAL IMPACT STATEMENT

  • Participants created storyboards from personal experiences, gave informed consent, and could skip questions, change topics, or withdraw without penalty.Researchers avoided prompts requiring highly sensitive or identifying disclosures.Source: ETHICAL IMPACT STATEMENT · p. 8
  • Drawings, interaction logs, generated references, and transcripts could indirectly reveal personal experiences or identities; the team removed identifying details and restricted data access.Published artifacts were screened to avoid revealing identifying information.Source: ETHICAL IMPACT STATEMENT · p. 8
  • The authors note that generative media can be misused for deceptive or non-consensual depictions and may reproduce training-data biases, including stereotyped portrayals.The work does not attempt to resolve these broader challenges.Source: ETHICAL IMPACT STATEMENT · p. 8
  • Findings may not generalize across cultures, and comparison with unassisted drawing may under-represent artists’ workflows involving reference gathering.The authors call for external assessments and a diverse, cross-cultural participant base.Source: ETHICAL IMPACT STATEMENT · p. 8
Loading 2610.09155v1…