Exploring the Potential of Generative AI in Expressing Brand Design Language in Product Design

Abstract

This study examines whether generative AI can consistently express Brand Design Language (BDL) in product design and how a brand-conditioned system differs from a generic generative AI model. A Custom AI was developed by integrating Stable Diffusion 1.5 with LoRA fine-tuning, IP-Adapter visual conditioning, and GPT-based prompt engineering. Using Apple, Nike, and Allurewave as target brands, the study evaluates eight internal control combinations and compares Custom AI outputs with Midjourney through expert interviews and a general-user survey. The Experiment shows that brand-likeness emerges most clearly when multiple control channels operate together, and that optimal A/B/C strategies vary by brand: Apple benefits from activating A/B/C together, Nike benefits from minimizing textual constraint while keeping B/C active, and Allurewave balances innovation and consistency by flexibly toggling C. Validation 1 indicates that experts can recognize brand-specific formal cues in Custom AI outputs and position the system as an early-stage ideation partner, while also noting limitations in finish realism for some cases. Validation 2 shows that general users preferred Custom AI for Nike (60.8%) and Allurewave (80.1%), driven mainly by form, mood, and color cues; for Apple, Midjourney was more often selected because it more consistently matched prototypical minimal cues (e.g., rounder radii and reduced detailing), with material/finish impressions mentioned only occasionally. Overall, the findings show that generative AI can express BDL when properly conditioned, but effectiveness depends on brand characteristics and evaluator perception. The study offers a framework for operationalizing BDL within generative systems and highlights the emerging role of designers as curators who define and refine brand design language in collaboration with AI.