AIdeation: Designing a Human-AI Collaborative Ideation System for Concept Designers

Research Background and Problem

• What problems or challenges did the authors identify?
  The authors identified several challenges faced by concept designers during the early ideation phase, including difficulties in gathering reference materials, generating diverse design variations, and working under extreme time pressures to complete tasks quickly. Additionally, current generative AI tools often lack support for complex design workflows and fail to align with designers' iterative processes.

• Why is this problem important?
  Concept designers play a critical role in the entertainment industry, transforming ideas into visual representations across film, gaming, and television. Constraints on creativity during the early ideation phase can impact the efficiency and quality of the entire design process, potentially leading to outcomes that fail to meet client expectations. With the rapid development of generative AI, better integration of AI into designers' workflows has become a key research direction.

• Research Motivation and Related Work
  Many existing AI tools struggle to support iterative and exploratory design practices (e.g., refinement and flexibility). Previous research has attempted to apply generative AI to fields such as graphic design, animation, and interior design, but there is still limited support for the highly specialized domain of concept design. Therefore, the authors aim to design a human-AI collaborative tool to address these challenges by systematically supporting the unique workflows of concept designers.

Solution

• What methods or solutions did the authors propose?
  The authors designed and implemented a human-AI collaborative ideation tool called AIdeation. This system enhances the efficiency and quality of concept designers' early creative processes by integrating multiple generative models and keyword extraction functionalities.

• What are the innovative aspects of this solution?
  AIdeation was designed with three key goals tailored to designers:

  • Supporting broad exploration by generating diverse design ideas through natural language input.
  • Providing in-depth research capabilities by extracting keywords for each design concept and linking reference resources to facilitate deeper exploration.
  • Enabling flexible iteration by integrating references or natural language instructions to refine designs and enhance creative control.

  Compared to existing AI tools, its core innovation lies in integrating research, brainstorming, and design refinement into a single system that aligns with designers' iterative workflows.

• What are the implementation steps and key technologies used?

  • Broad Exploration (Brainstorming Phase): Based on user input, the system uses the GPT model to generate multiple design concepts and employs DALL-E 3 to create visual images. The generated results are categorized by themes, content, art styles, lighting, and atmosphere.
  • In-Depth Exploration (Research Phase): Keywords are extracted using GPT, and linked to real-world image search engines (e.g., Bing Image Search) to provide authentic reference images.
  • Flexible Iteration (Refinement Phase): By combining reference images or natural language instructions, the system generates new design variations, enabling users to expand on existing designs or focus on specific design elements.

Research Outcomes

• What specific outcomes were achieved?

  • In a comparative experiment involving 16 professional environment concept designers, AIdeation significantly improved creative efficiency, the ability to generate diverse design concepts, and participants' satisfaction and sense of task accomplishment.
  • During a one-week field study in real-world commercial projects, four design studios reported practical benefits of the tool in enhancing creativity and quality, with two studios continuing to use AIdeation after the experiment.

• What advantages does it have compared to existing solutions?

  • AIdeation significantly enhances both broad and deep exploration capabilities while supporting flexible iteration in designers' workflows.
  • Compared to single-step generative tools like MidJourney and DALL-E, AIdeation offers greater design controllability and integrates task-relevant information.
  • In addition to generating visual images, it provides verifiable reference information through keyword extraction, effectively addressing the "AI hallucination" problem in generated content.

• What were the experimental or evaluation results?

  • In the comparative experiment, participants rated AIdeation as superior to their original workflows in enhancing creativity (p = 0.001), task efficiency (p = 0.003), and satisfaction (p = 0.005).
  • In field testing, some studios reported a 25%-60% reduction in project completion time while improving design quality.

• Limitations and Future Directions

  • Limitations: The current system's control features are relatively basic, lacking precise adjustments for visual details such as lighting and composition; the generated content's artistic styles remain somewhat limited.
  • Future Directions:
    • Enhance detailed control over generated results, such as adding inpainting functionality.
    • Provide multiple generative model options (e.g., MidJourney or Stable Diffusion) to further support diverse artistic styles.
    • Develop personalized and adaptive features to dynamically optimize feedback based on user preferences and task stages, and expand to other design domains such as industrial design and fashion design.

Through this analysis, it is evident that AIdeation successfully addresses the critical needs of concept designers during the early ideation phase, offering a new perspective on the application of AI tools in the design field.