DesignTrace: Exploring, Iterating and Tracking Design Alternatives with GenAI

Paper Title

DesignTrace: Exploring, Iterating and Tracking Design Alternatives with GenAI

Publication Info

• Topic area: GenAI-based creativity support tools for design workflows
• Keywords: GenAI, design alternatives, iterative design, visual consistency, design history, node-based interface, semantic attributes, creative workflows, human-AI interaction, design rationale

Background and Problem

• Problem / challenge: Current GenAI tools lack explicit support for iterative evolution, reflection, and tracking of design alternatives. Designers struggle with fragmented tools, inconsistent outputs, and limited control over design history and provenance.
• Significance: Addressing these gaps is crucial for enabling professional designers to explore and refine creative ideas while maintaining control and consistency, especially as GenAI becomes integral to design workflows.
• Motivation and related work: Prior systems focus on rapid image generation or static alternatives but fail to support iterative exploration and refinement. Existing tools often prioritize technical expertise, making them inaccessible to visually-oriented designers. There is a need for systems that integrate GenAI capabilities with intuitive, visual workflows.

Solution

• Proposed approach: DesignTrace, a node-based GenAI design tool that enables designers to explore, iterate, and track design alternatives through semantic attributes, interactive nodes, and structured layouts.
• Novelty:
  1. Reifying design rationale into manipulable artifacts with semantic and visual elements.
  2. Introducing a node-based interface for iterative exploration and visual history tracking.
  3. Supporting multi-scale traces combining command histories, artifact states, and exploration paths.
  4. Empirical evaluation of DesignTrace with professional designers, highlighting its advantages over baseline tools.
• Procedure and key techniques:
  • Semantic attributes are extracted, created, and stored in a Palette for reuse.
  • Nodes act as interactive units for visual editing and history tracking, enabling branching and localized edits.
  • Templates allow customization of GenAI prompts for structured transformations.
  • Multi-scale traces capture design progress across temporal and spatial dimensions, supporting backtracking and reuse.

Results

• Concrete findings:
  • 75% of participants preferred DesignTrace for customizing and reusing assets, tracking progress, and organizing design choices.
  • Designers created an average of 5 categories and 9 attributes in the Palette, reusing attributes 1–4 times.
  • Designers restarted explorations on earlier Nodes 10/12 times (mean: 2), compared to 5/12 times (mean: 0.58) in the baseline condition.
• Advantage over baselines:
  • DesignTrace facilitated visual consistency across iterations, better organization of design artifacts, and easier backtracking compared to baseline tools.
  • The Palette encouraged design reflection and intentionality, while Nodes supported structured workflows and reuse.
• Experiments / evaluation:
  • Comparative Structured Observation study with 12 professional designers using two design briefs.
  • Participants performed design tasks under two conditions (node condition vs. baseline) and provided qualitative and quantitative feedback.
• Limitations and future work:
  • Limited study time constrained usage of advanced features like Template customization.
  • Scalability issues with crowded interfaces after extended use.
  • GenAI outputs occasionally deviated from designer intent due to model limitations.
  • Future work includes hierarchical trace management, improved predictability of attribute-driven transformations, and integration with existing design tools.

Summary

DesignTrace introduces a node-based interface that reifies design rationale into semantic and visual artifacts for iterative exploration and tracking of GenAI-generated alternatives. Empirical evaluation with 12 professional designers demonstrated its advantages in maintaining visual consistency, supporting design reflection, and organizing spatial-temporal histories. The Palette, Nodes, and Template collectively enabled designers to create intentional assets, explore alternatives systematically, and reuse design elements effectively. While scalability and GenAI unpredictability remain challenges, DesignTrace provides a promising framework for integrating GenAI into professional visual design workflows.