Researchers at the intersection of artificial intelligence and fashion design have proposed a new branding framework intended to assist early-stage designer brands in navigating the complexities of market positioning. The study, published in the journal Fashion and Textiles, introduces an AI-driven platform that integrates the Persona/Scenario (P/S) methodology with advanced machine learning components. The primary objective is to provide structured, data-informed support for fashion startups that often lack the resources for extensive consumer research and strategic planning. By combining visual style classification with semantic analysis of consumer profiles, the framework aims to bridge the gap between creative intuition and systematic brand development.
The research addresses a specific gap in the current fashion industry, where designer-led startups frequently rely on intuition rather than robust data infrastructures. According to the authors, many of these brands face significant challenges in brand development, including limited resources and insufficient understanding of their target consumers. Existing branding frameworks are often designed for large, resource-rich enterprises, leaving small-scale operations with limited relevance. The proposed system seeks to offer a scalable and flexible pathway for these emerging brands to achieve systematic and consumer-oriented growth.
The technical architecture of the platform relies on three core AI modules: a Style Analysis Module, a Text Processing Module, and a DeepSeek Module. The Style Analysis Module utilizes a Mixture-of-Experts Convolutional Neural Network (MoE-CNN) to classify fashion images into multiple aesthetic categories. Unlike traditional CNNs, the MoE-CNN employs a gating mechanism to assign input samples to different expert subnetworks, each focusing on specific style attributes such as color, pattern, and silhouette. This structure is particularly effective for designer brands whose garments often present hybrid styles combining diverse visual elements.
To ensure accurate classification, the system includes a preprocessing step that segments garments from backgrounds and human figures using SAM2-based image segmentation. The model was trained on a curated dataset of 29,871 images collected from 70 representative designer brands in China and South Korea. These images were sourced from digital platforms such as Musinsa, W Concept, Xiaohongshu, and Taobao, covering collections from 2024 to 2025. Style annotations were manually labeled by five fashion design researchers based on industry trend reports and mainstream fashion classification systems to ensure professional relevance and consistency.
The Text Processing Module is designed to interpret the designer’s conceptual input and infer appropriate user persona suggestions. It computes semantic embeddings of the designer’s textual input and compares them against a curated persona database using a K-nearest Neighbors (KNN) algorithm. This database comprises both quantitative and qualitative data sources, including survey responses from 322 fashion consumers aged 20 to 30 in China and South Korea, as well as behavioral data mined from Instagram and Xiaohongshu. Each persona entry is structured around lifestyle values, aesthetic preferences, functional needs, and emotional motivations, bridging the gap between internal brand identity and external user targeting.
The final component, the DeepSeek Module, translates persona and scenario structures into actionable brand strategies using the DeepSeek-VL vision-language model. This module generates candidate user personas, enriches them with detailed behavioral and psychological attributes, and produces brand-building insights such as trend forecasts and style positioning directions. It also simulates user-brand interactions across various touchpoints and lifestyle contexts, allowing designers to visualize real-world engagement. The system synthesizes these scenarios into targeted marketing strategies, offering emotional triggers, content themes, and channel recommendations to support cohesive brand messaging.
The development of this framework was grounded in qualitative research involving in-depth interviews with ten fashion designers, including six Chinese and four Korean designers. These participants were selected for their recognized market presence and experience, with their brands featured on prominent digital platforms. The interviews were conducted via Zoom and lasted approximately 90 minutes each. Transcripts were analyzed using thematic coding, resulting in 1,719 individual codes clustered into 177 initial categories, which were further grouped into 45 subcategories and ultimately 10 core themes. This process provided empirical grounding for the identification of branding challenges faced by startup fashion designer brands.
Key themes identified in the interview analysis included motivations for brand creation, early brand management challenges, and the reliance on intuition-driven planning due to financial constraints. Designers reported difficulties in accessing investors, conducting structured consumer research, and managing operational overload. Many relied on informal feedback and personal networks to define target audiences, leading to vague positioning. The study also highlighted the importance of brand distinctiveness, with designers prioritizing style, philosophy, and consumer identity over price or scale. Competitive benchmarking via peer observation on social media platforms served as a primary research method in the absence of structured tools.
To evaluate the proposed platform, a usability test was conducted with five fashion designers who had previously participated in the initial interview phase. The evaluation utilized a structured questionnaire based on the Technology Acceptance Model (TAM), measuring Perceived Usefulness (PU), Perceived Ease of Use (PEOU), and Behavioral Intention (BI). Quantitative results indicated a generally high level of user acceptance, with result demonstrability and job relevance rated at 4.66 and 4.60, respectively, on a five-point Likert scale. Perceived ease of use received the highest scores, with perceived enjoyment and computer self-efficacy rated at 4.66 and 4.74, reflecting strong user engagement and confidence in navigating the system.
Participants also evaluated seven branding workflow stages, revealing a progressive increase in user satisfaction across these stages. Style recognition, the first stage, received the lowest score at 4.23, suggesting a need for improved interpretability of the system’s visual analysis outputs. In contrast, market positioning and strategic keyword discovery achieved the highest score of 4.87. Scenario generation and marketing suggestion output also received strong ratings of 4.73 and 4.70, respectively. These findings indicate that participants perceived the strategic output functions of the system to be particularly useful and relevant to their branding needs.
Qualitative feedback from the usability test highlighted both strengths and areas for improvement. Designers appreciated the platform’s modular clarity and visualized process flow, noting that it effectively supported tasks that were traditionally time-consuming or lacked methodological structure. However, a recurring concern was the uncertainty regarding the accuracy of AI-generated content, particularly with respect to market trend suggestions. Some designers acknowledged that they lacked a clear understanding of current fashion trends themselves, making it difficult to judge the validity of AI-generated insights. This highlights the need for greater transparency and explainability in future iterations of the system.
The study concludes that the integration of AI and the P/S methodology offers a strategic value for startup fashion branding, shifting brand conception from intuition-driven ideation toward a more evidence-based and user-informed strategic logic. The platform addresses a critical gap in current fashion AI ecosystems, which largely emphasize operational tasks while lacking integrated, strategy-oriented support for brand development. Despite its strengths, the study acknowledges limitations in AI transparency and the need for further validation across subsequent lifecycle phases, including growth, maintenance, and expansion. Future research should incorporate more heterogeneous datasets and enhance algorithmic sensitivity to cultural variation to ensure global scalability.
By providing a structured yet adaptable decision-support system, this research contributes to the interdisciplinary integration of AI technologies and fashion branding. It enables a strategic branding path that fuses user orientation with design-driven thinking, offering a potential tool for designers to navigate the rapidly evolving fashion market. The platform’s regional adaptability, validated through localized user data and culturally responsive inference logic, supports its contextual alignment in both Chinese and Korean markets, demonstrating its potential as a scalable and adaptive branding infrastructure.
Subject of Research: Development of an AI-driven branding platform integrating persona/scenario methods for fashion startups
Article Title: Development of an AI-driven branding platform integrating persona/scenario methods for fashion startups
Article References: Ma, N., & Lee, J. H. (2026). Development of an AI-driven branding platform integrating persona/scenario methods for fashion startups. Fashion and Textiles, 13(1), Article 11. https://doi.org/10.1186/s40691-026-00461-2
Image Credits: AI Generated
DOI: 10.1186/s40691-026-00461-2
Keywords: Development, AI-driven, branding, platform, integrating, persona, scenario, methods, fashion, startups, scientific research
News Source: Denise Maddox. (November 8, 2026). AI Platform Integrates Persona Methods for Fashion Startups. Scienmag.
