CharacterMeet: Supporting Creative Writers' Entire Story Character Construction Processes Through Conversation with LLM-Powered Chatbot Avatars

Title of the Paper

CharacterMeet: Supporting Creative Writers’ Entire Story Character Construction Processes Through Conversation with LLM-Powered Chatbot Avatars

Paper Information

• Subject Area: Creative Writing and Human-Computer Interaction
• Keywords: Creativity Support, Writing Assistance, Creative Writing, AI Collaboration, Large Language Models

Research Background and Issues

• Identified Problems or Challenges:

  1. Character construction is a crucial part of creative writing, but there is a lack of tools that support the entire character construction process, especially those that integrate conceptualization and expression stages.
  2. Existing technologies (e.g., templates, traditional dialogue tools) have limited flexibility in inspiring creators and struggle to provide immersive or personalized support.
  3. LLM-based support systems (e.g., GPT-4) have not been fully explored in terms of enhancing the character construction process through real-time dialogue and visualization.

• Significance of the Research: Providing robust creativity support tools can enhance writers' productivity, inspire creativity, and enrich methods of character expression. Such tools are in high demand in creative education and professional writing fields.

• Motivation and Related Work: Previous research (e.g., CharacterChat) introduced the concept of "progressive rendering," which transforms characters gradually through dialogue but is limited to the initial conceptualization stage. Advances in generative AI and multimodal technologies offer new possibilities for deeper and more personalized character construction support.

Solution

• Proposed Method or Solution: CharacterMeet is a prototype system that combines generative AI (GPT-4), 3D character animation, and text-to-image generation technologies (e.g., Stable Diffusion) to support complete character construction through dialogue and personalized visualization tools.

  • Users can iteratively update character attributes, backstories, and dialogue contexts, while visualizing them through background images and character appearances.

• Innovative Aspects:

  1. Integration of advanced LLM and 3D avatar technologies to support progressive rendering and multimodal visualization of characters.
  2. Support for the entire character construction process, encompassing both conceptualization (ideation) and expression (implementation through writing).
  3. Personalized design catering to different creators' sources of inspiration and preferences, such as text, audio, and visual stimuli.

• Implementation Steps and Key Technologies:

  • The system generates dialogues using OpenAI GPT-4, background images via Blockade Labs' Skybox AI, and character appearance customization through the Ready Player Me platform.
  • Users can input free text to describe character attributes and backstories, with support for voice output of generated content.

Research Outcomes

• Specific Results:

  1. A prototype tool (CharacterMeet) was proposed to support complete character construction through character dialogue.
  2. User experiments (14 creators) demonstrated that the system supports phased, iterative character construction while meeting personalized creative needs.
  3. Creators could switch between the writing process and character exploration, generating new inspiration for dialogues, backstories, and character personalities through interactions.

• Comparison with Existing Solutions:

  • Compared to solutions that only support the initial conceptualization stage (e.g., CharacterChat), CharacterMeet covers the entire process from conceptualization to expression.
  • Unlike static recommendation or auto-generation tools, this system allows users to explore dynamic character reactions in specific contexts, enhancing immersion.

• Experimental or Evaluation Results:

  1. The system scored 80.02/100 on the "Creativity Support Index (CSI)," higher than predecessor tools in related studies (score: 76.96).
  2. The system received high ratings in exploration, immersion, and expressiveness but showed slight shortcomings in information tracking functionality.
  3. User feedback highlighted the positive impact of personalized audio and visual effects on improving the creative experience.

• Limitations and Future Directions:

  1. Limitations:

     • Small sample size and cultural-linguistic diversity introduced complexity in evaluation.
     • Quality assessment of system output relied on subjective user perceptions, lacking objective standards.
     • The study focused only on human-like characters, neglecting other possible forms (e.g., animals, non-human entities).

  2. Future Directions:

     • Develop personalized tools applicable to multilingual and multicultural contexts for the entire writing process.
     • Explore the tool's application in other narrative forms (e.g., games or screenwriting).
     • Investigate the impact of visualization on creators' imagination and explore more dynamic visual presentation options (e.g., multi-character interaction).

Conclusion

CharacterMeet demonstrates the potential of character construction tools based on dialogue and multimodal generation technologies in creative writing. Future research could further explore its ability to support the complete writing process and its applicability in multicultural and cross-linguistic creative environments.