DesignLab: Designing Slides Through Iterative Detection and Correction

Designing high-quality presentation slides can be challenging for non-expertsdue to the complexity involved in navigating various design choices. Numerousautomated tools can suggest layouts and color schemes, yet often lack theability to refine their own output, which is a key aspect in real-worldworkflows. We propose DesignLab, which separates the design process into tworoles, the design reviewer, who identifies design-related issues, and thedesign contributor who corrects them. This decomposition enables an iterativeloop where the reviewer continuously detects issues and the contributorcorrects them, allowing a draft to be further polished with each iteration,reaching qualities that were unattainable. We fine-tune large language modelsfor these roles and simulate intermediate drafts by introducing controlledperturbations, enabling the design reviewer learn design errors and thecontributor learn how to fix them. Our experiments show that DesignLaboutperforms existing design-generation methods, including a commercial tool, byembracing the iterative nature of designing which can result in polished,professional slides.