Style AI

𝙎𝙩𝙮𝙡𝙚 𝘼𝙄 is an innovative system that generates dresses through an interactive genetic algorithm. This service creates a dress for a specific occasion and style selected by the user. By answering questions and selecting their favourite dresses in each population, the user is guiding the algorithm into creating the perfect dress.

Timeline

2023

Project Course (MSc)

Computational Creativity @ Leiden University

Role (3 Members Team)

Research: Background & market analysis
Flowchart & Wireframing
Branding
Prototyping (Hi-fi)

project overview

Generative AI (agents) has been actively used for various purposes. Through this project, the team dived deeper into the followings:

  1. show AI-based services are being developed

  2. the definition of (computational) creativity

  3. the process of co-creation between human users and the agents

Design challenge

Design an AI-powered co-creation platform for a custom clothing

Fashion is Subjective

In this project, we were asked to use a genetic algorithm and user input. At the same time, we noticed that fashion is subjective that it is almost impossible to measure as a value; the opinion of the user is crucial for the algorithm to create the ideal dress. This led to a question- How would an experience be like when AI and human creativity work together for a custom clothing design?

My Role as UX Researcher & Design Strategist

UX Research & Ideation

I conducted background and market research to understand how AI was currently being applied in the fashion sector, with particular attention to user co-creation processes. Drawing from academic literature and competitive analysis, I defined key user needs and preferences for our system.

System Design & Flow Mapping

I led the creation of system flowcharts to ensure a smooth and logical interaction between user input and algorithmic output. This included defining how users influence the generative process by selecting favorite designs from each population.

Interface Design & Prototyping

I developed high-fidelity wireframes and interactive prototypes in Figma to visualize the system's experience. I ensured the branding and UI reflected the creativity and personalization central to the project’s goals.

research

Exploratory Research

We studied current use cases of AI in fashion, focusing on computational creativity and the implications of AI-human co-creation. Our target users were fashion-conscious women interested in custom dress design-users who value both aesthetic control and innovation.

Results: Literature Review

We found two primary ways AI can participate in fashion design:

  1. Mix-and-match styling: AI suggests combinations from users’ existing wardrobe items.

  2. New item generation: AI designs entirely new clothing items beyond the user's current wardrobe.

To meaningfully explore computational creativity, we focused on ‘New item generation’, enabling AI to create new and original fashion pieces. This approach presents a clearer opportunity to evaluate AI's ability to go beyond replication and engage in generative design.

Evaluating two approaches

  • Jiang & Fu (2017) proposed a method that merges an artwork or design image with an existing clothing item to generate a new piece. While visually appealing, we identified key limitations:

    • Heavy reliance on user-provided inputs restricts the creative range.

    • Output designs risk plagiarism by borrowing too closely from original artworks.

    • The system's role is more decorative than generative.

    Jiang, S., & Fu, Y.R. (2017). Fashion Style Generator. International Joint Conference on Artificial Intelligence.

  • Dubey (2020) introduced a more flexible system that blends features from multiple clothing items to synthesize a new design. This approach aligned with our goals for three main reasons:

    • It enables originality by moving beyond full-pattern transfer.

    • It reduces the dependency on user-supplied images, allowing more autonomous generation.

    • It introduces space for exploration, as AI infers user preferences and creates designs from combinations of abstracted features.

    Dubey, A., Bhardwaj, N., Abhinav, K., Kuriakose, M.S., Jain, S., & Arora, V. (2020). AI Assisted Apparel Design. ArXiv, abs/2007.04950.


ideation

We adopted the feature recombination model as our foundational concept. This direction allows for higher novelty, reduces plagiarism concerns, and invites meaningful human-AI co-creation by integrating user preferences into an evolving design process. It became the core framework for both our algorithmic logic and UX strategy.

The following shows the flowchart of the experience.

Final Design

update the design accordingly

The system was powered by a Python-based genetic algorithm that iteratively refined dress options based on user choices. The interface and interaction flows were prototyped using Figma to simulate the user experience.

The design involves three main steps- 1) choose the occasion, 2) add preferences along the way & the algorithm updates accordingly, and 3) edit the details. Based on the user input, the AI algorithm is being updated.

check the prototype

reflection

We developed our assessment metrics inspired by other researchers. The metrics are:

  • Independence

  • Variety and experimentation

  • Originality

  • Exploration

  • Expressiveness

  • Enjoyment

  • Collaboration

  • Satisfaction with final outcome

    • Cherry, E., & Latulipe, C. (2014). Quantifying the Creativity Support of Digital Tools through the Creativity Support Index. ACM Transactions on Computer-Human Interaction, 21(4), 1–25. doi: 10.1145/2617588

    • Jordanous, A. A. (2012). Standardised Procedure for Evaluating Creative Systems: Computational Creativity Evaluation Based on What it is to be Creative. Cognitive Computation 4, 246–279. doi: 10.1007/s12559-012-9156-1

    • Kantosalo, A., & Riihiaho, S. (2018). Experience evaluations for human–computer co-creative processes – planning and conducting an evaluation in practice. Connection Science, 31(1), 60–81. doi: 10.1080/09540091.2018.1432566

From the evaluation, we identified positive aspects of the designed experience. Due to the small sample size (n=5), all statistical interpretations should be considered exploratory and preliminary.

4.2 / 5

overall satisfaction

  • Easy and enjoyable collaboration process

  • Satisfying design outcomes

  • High intention to reuse the system

The outcome remains areas for improvement, regarding 1) limited novelty in design outputs and 2) moderate user independence in guiding the process.

takeaways

Reflection on Co-Creation & Evaluation Strategy
While the project successfully demonstrated co-creation between users and the AI system, the small sample size limited the generalisability of the findings and the depth of insight into user behaviour.

→ Increase participant diversity and establish clearer metrics for evaluating creativity and user agency in future iterations.

Clarifying System Influence on Creativity
The genetic algorithm provided a good foundation for co-creative output, but the system’s influence often overshadowed user independence during decision-making.

→ Refine the user interaction model to better balance algorithmic suggestions with user-led exploration.

Dataset and Output Diversity
The limited richness of the training data resulted in outputs that occasionally lacked originality and variation.

→ Explore ways to diversify input sources and implement dynamic data augmentation to enhance the novelty of generated outcomes.

Next: BeachBot