LumiMood: A Creativity Support Tool for Designing the Mood of a 3D Scene

Document Title

LumiMood: A Creativity Support Tool for Designing the Mood of a 3D Scene

Document Information

• Research Domain: Studies on game design and creativity support tools, particularly AI-driven emotional and visual design.
• Keywords: Creativity support tools, affective computing, graphic design, artificial intelligence, game scenes, mood design, lighting design, post-processing, AI-generated content, visual emotion analysis.

Research Background and Problem

• Key Issues and Challenges:

  • In game scene design, creating immersive and emotionally evocative atmospheres is a primary task, requiring detailed design of low-level features such as color distribution, contrast, and brightness.
  • Current AI support tools mainly focus on the automatic generation of semantic properties for 3D models, with limited automation support for the overall low-level features of scenes.
  • Challenges in mood design include translating abstract concepts into visual representations, the tedious workflow of trial-and-error for setting and evaluating attribute values, and the complexity of learning and understanding attributes for precise adjustments.

• Significance of Research:

  • The study aims to address major design challenges in creating atmospheres for game scenes by introducing AI technologies to reduce design time, improve accuracy, and provide designers with more convenient tools.
  • Mood design is critical for the user experience of games, influencing players' emotional responses, immersion, and the overall quality of the game.

• Motivation and Related Work:

  • AI technologies, such as Generative Adversarial Networks (GANs) and diffusion models, have profoundly impacted fields like image generation and 3D modeling. Combined with natural language processing methods, AI can effectively simplify complex design tasks.
  • Existing research predominantly focuses on semantic properties, with limited work addressing mood design that simultaneously considers low-level visual features and emotions.
  • This study aims to introduce new workflows and tools to provide comprehensive support for designers, particularly in adjusting lighting and post-processing to achieve emotional design goals.

Solution

• Main Methods and Tools:

  • Introducing LumiMood, an AI-based creativity support tool specifically designed for mood design in 3D scenes.
  • LumiMood consists of three main components:
    1. Reference Image Generator (Generator): Generates emotional reference images using diffusion models.
    2. Template Scene Creator (Creator): Automatically adjusts lighting and post-processing attributes to produce template designs.
    3. Design Step Recaller (Recaller): Tracks design steps and demonstrates how attribute adjustments affect scene appearance.

• Innovations:

  • Utilizing diffusion models to generate emotional reference images, aiding designers in translating abstract mood concepts into visual representations.
  • Providing automated template designs to simplify the tedious trial-and-error process.
  • Displaying intermediate design steps to enhance designers' understanding and control of attributes.

• Implementation Steps and Key Technologies:

  • The Generator component employs a diffusion model trained on the WEBEmo dataset to generate reference images based on emotional keywords.
  • The Creator implements a grid search algorithm, optimizing lighting and post-processing attributes in two stages to minimize the mean squared error between the designed scene and the reference image.
  • The Recaller records and traces adjustment steps, calculates the impact scores of each attribute on the design outcome, and ranks and displays the most significant attribute adjustments.

Research Outcomes

• Specific Results:

  • LumiMood significantly improves design efficiency: reducing design time and interaction frequency while enhancing design accuracy and content satisfaction.
  • Both professional and novice designers benefit from its functionalities, though differences in usage styles were observed.
  • User studies validate that LumiMood helps designers address the challenge of translating abstract mood concepts, reduces tedious design workflows, and deepens understanding of attributes.

• Comparison with Existing Solutions:

  • LumiMood supports not only high-level content design but also provides comprehensive support for low-level features.
  • Compared to directly using game engines, designers can complete tasks more quickly and achieve significantly higher design precision.

• Experiment and Evaluation Results:

  • User experiment: 40 designers participated in tasks, including scene replication and mood design tasks.
  • Results indicate:
    1. Design time decreased by approximately 30%-40% with LumiMood.
    2. Professional designers demonstrated better understanding and application of lighting and post-processing attributes.
    3. LumiMood was particularly beneficial for novice designers, providing clear directions through automated generation.

• Limitations and Future Directions:

  • The diffusion model used in the Generator is a black-box mechanism lacking transparency; future research could explore explainable AI technologies.
  • Enhancing the flexibility of the Generator to support cross-scene style consistency.
  • Expanding support for high-level features (e.g., 3D models and semantic content) to further enhance design freedom and diversity, avoiding the issue of rigid generated content.
  • Customizing the dynamic balance between automation and control based on user expertise levels to meet the needs of both beginners and professional designers.

Conclusion

By deeply exploring the needs and challenges of designers, LumiMood brings significant support and improvements to game scene design. It organically integrates AI models with design workflows, providing valuable references for the design of creativity support tools.