We-toon: A Communication Support System between Writers and Artists in Collaborative Webtoon Sketch Revision

Title of the Paper

We-toon: A Communication Support System between Writers and Artists in Collaborative Webtoon Sketch Revision

Paper Information

• Research Area: Human-Computer Interaction, Collaborative Systems, Applications of Generative AI in Artistic Creation
• Keywords: Webtoon, Collaboration, Communication, Human-Computer Interaction, AI-Assisted Creativity, Interactive Systems, User Interface, GAN, User Study, Usability Study

Research Background and Problem Statement

• Identified Problems and Challenges:

  1. In the creation process of Webtoons, effective communication between writers and artists is essential, especially during character design and sketch revision. However, writers often struggle to accurately describe drawing details through text, leading to repeated revisions.
  2. Existing methods mainly rely on textual descriptions and image searches (e.g., Google, Pinterest) to provide references, which are inefficient and unintuitive.
  3. There is a lack of tools specifically designed for the Webtoon creation domain to simplify communication and improve collaboration efficiency.

• Significance of the Research:

  • With the widespread use of mobile devices, the Webtoon industry has grown rapidly, making its collaborative environment increasingly complex and systematic. Improving communication efficiency between collaborators directly impacts the quality and publishing efficiency of the work.

• Research Motivation and Related Work:

  • Based on preliminary interviews with four professional writers, this study identifies the need to enhance collaboration efficiency. By integrating cutting-edge technologies (e.g., Generative Adversarial Networks, GANs), the study aims to address issues of "ambiguous feedback" and "revision efficiency."
  • Related literature focuses on multi-user online collaborative systems and image generation technologies but does not address the specific context of Webtoon creation.

Solution

• Method and System Overview:

  • The study proposes the We-toon system: a GAN-based communication support tool that helps writers generate reference images and convey modification intentions to artists.
  • The system employs a two-stage process:
    1. Image Preparation: Generating or selecting predefined reference images based on user-defined attributes.
    2. Image Synthesis: Integrating specific areas of the reference image into the artist's original sketch.

• Innovations:

  1. Utilizes StyleGAN2-Ada and StyleMapGAN to generate high-quality reference images and supports localized image editing.
  2. Combines textual descriptions, hand-drawn annotations, and high-intelligence synthesized images to enhance communication intuitiveness.
  3. Introduces interactive features such as "attribute filtering" and "fine-tuning," enabling writers to quickly generate reference materials without requiring professional drawing skills.

• Implementation Steps and Key Technologies:

  1. Image Generation and Selection:
     • Built a dataset of 47,233 Webtoon character facial images and used StyleGAN2-Ada to generate unlimited new images.
     • Writers can adjust image attributes (e.g., gender, hairstyle, eye size) through various interactive methods.
  2. Image Synthesis:
     • Utilized StyleMapGAN for localized style transfer, allowing users to define "modification content" and "modification location."
  3. User Interface Design:
     • Designed a dedicated interface for writers to convey "revision requests," including text instructions, annotated areas, and synthesized images.

Research Outcomes

• Specific Results:

  • User experiments involving 24 professional writers demonstrated that We-toon significantly improves communication efficiency between collaborators.
  • Writers reported higher satisfaction with the appearance, harmony, and detail representation of images generated by We-toon compared to traditional methods.

• Comparative Advantages over Existing Solutions:

  1. Improved Communication Clarity:
     • The generated images are intuitive and specific, addressing the ambiguity and lack of clarity in text-only descriptions.
     • Reduces reliance on image searches, enabling the generation of reference images that meet expectations without external resources.
  2. Enhanced Time Efficiency:
     • We-toon significantly reduces the number of revision requests and the time spent on each revision (average reduced to 4.6 minutes).
  3. Increased User Satisfaction:
     • Both writers and artists agreed that the system improved mutual understanding and provided better support for the final output.

• Experiment and Evaluation Results:

  • Key Findings:
    • 86% of writers believed that We-toon could generate diverse and high-quality reference images.
    • 83% of artists reported that revision requests based on We-toon were easier to understand than traditional methods.
    • Scores for appearance, harmony, and detail completeness improved by an average of 12% compared to traditional methods (Baseline).
  • Limitations and Areas for Improvement:
    1. Currently supports only facial close-ups and fixed expressions; expansion to full-body characters and dynamic expressions is needed.
    2. The system does not yet support global semantic modifications; future work could incorporate more advanced models (e.g., HyperStyle) to enhance flexibility.
    3. Attribute classification still requires manual input; plans for future development include automation through self-supervised learning.

• Future Directions:

  1. Explore the potential applications of We-toon in other art forms (e.g., animation, illustration design).
  2. Develop more versatile GAN modules for complex scenes and dynamic objects.
  3. Investigate copyright and legal issues related to AI-generated content.