Agentic Design Review System

Abstract

Evaluating graphic designs involves assessing it from multiple facets like alignment, composition, aesthetics and color choices. Evaluating designs in a holistic way involves aggregating feedback from individual expert reviewers. Towards this, we propose an Agentic Design Review System (AgenticDRS), where multiple agents collaboratively analyze a design, orchestrated by a meta-agent. A novel in-context exemplar selection approach based on graph matching and a unique prompt expansion method plays central role towards making each agent design aware. Towards evaluating this framework, we propose DRS-BENCH benchmark. Thorough experimental evaluation against state-of-the-art baselines adapted to the problem setup, backed-up with critical ablation experiments brings out the efficacy of Agentic-DRS in evaluating graphic designs and generating actionable feedback. We hope that this work will attract attention to this pragmatic, yet under-explored research direction.

Figures (9)

• Figure 1: Evaluating a graphic design involves assessing it across multiple dimensions like visual coherence, semantic grouping, typographic clarity and so on. Inspired by peer-review system in conferences, we propose to build the first Agentic framework for design evaluation and feedback generation. We propose a novel approach to infuse design knowledge into agents, and allow them to collaboratively review and access the input design in \ref{['sec:method']}.
• Figure 2: Overview of the proposed design evaluation pipeline. To systematically evaluate designs, GRAD constructs a graph representation of design elements using edge and node matching techniques, enabling structured retrieval of in-context design examples from a curated library (Sec \ref{['sec:grad']}). Structured Design Description (SDD) module generates design descriptions (Sec \ref{['sec:sdd']}) to anchor the responses of each agent. The selected $K$ designs, query design along with its description inform the review process, where a meta agent coordinates static and dynamic agents to assess disparate design attributes - part of the Agentic Design Review System. Static agents focus on a fixed set of attributes (e.g., typography), whereas dynamic agents evaluate the attributes which are contextualized to specific designs (e.g., stylistic qualities). The meta agent consolidates these insights into a final rating and provides actionable feedback for design improvements (Sec \ref{['sec:adrs']}).
• Figure 3: Generated feedback along with the design attributes which are found to be inconsistent for the input design evaluated (best viewed when zoomed). Feedback evaluation scores are also reported (as outlined in Sec \ref{['sec:metrics']}).
• Figure 4: Impact of $K$ in GRAD for Agentic-DRS.
• Figure 5: Generated feedback along with the design attributes.