So, people are getting into using really smart computer tech, like artificial intelligence (AI), to create clothing. Some tech companies have developed AI systems using StyleGAN2 to change images, but they aren’t widely used in the fashion world yet. Fashion brands have to come up with lots of new designs each season, and using AI could speed up this process. This study is diving into existing AI tools for designing clothes and figuring out how they measure up to human methods. The goal is to create a new AI system that operates more like real fashion designers.
Table of Contents
Introduction: The Rise of AI in Fashion
AI is making waves, transforming various industries, fashion included. It helps online stores understand your preferences and predicts upcoming trends. Cool AI tools like Chat GPT and DALL-E2 can even generate images and more.
However, despite fashion’s rapid changes, designers still mostly trust their instincts. AI hasn’t completely taken over in fashion yet. This study suggests it’s because AI needs to grasp how actual designers work and stay updated on fashion trends to be genuinely useful.
Objectives of the Research
This study aims to do two things: firstly, examine existing AI tools for creating clothes and compare them to human methods. Secondly, utilize this information to develop a new AI system that comprehends human designers’ processes and is aware of fashion trends. The idea is to create AI tools that are more pragmatic and beneficial for real fashion tasks.
What’s the Plan?
The study is broken down into a few steps. Initially, they examine past cases where AI was applied in fashion. Then, they compare how these AI tools function compared to how human designers craft clothes. With this information, they aim to construct a new AI system using StyleGAN2. Finally, they’ll test it with actual industry designers to see if it’s genuinely something they would find useful.
Literature Review: How Clothes Are Made and What AI Knows
Creating clothes is like cracking a puzzle for designers. They have to consider the brand, the season, and people’s preferences. According to this study, five steps are crucial for fashion brands: data analysis, selecting the season’s theme, designing, refining designs, and finalizing them.
Fashion brands start planning their clothing well before it hits the stores. They analyze past sales and global trends, then plan the season’s theme, setting the tone for the designs. The actual design process involves creating and adjusting designs that align with the brand and season. Finally, the designs undergo reviews by merchandisers and others to determine what goes on sale.
The study emphasizes “domain knowledge,” which is like specialized knowledge in the fashion world, encompassing brand identity, past sales, and consumer preferences. This kind of knowledge is crucial in fashion, where human intuition plays a significant role.
AI Tech in Fashion Design
AI in fashion has seen significant evolution. In the early 2000s, genetic algorithms were used to suggest new styles based on existing ones. Later, computer vision was employed to recognize items in photos and understand design attributes. Now, GAN (Generative Adversarial Network) models are making waves. They use GAN to generate new designs, modify existing ones, and even create graphics for clothes. StyleGAN and StyleGAN2 are special algorithms for fashion images, offering control over styles while generating images.
So, the study aims to blend the creativity of human designers with the smart tech of AI, creating a system that genuinely suits the needs of the fashion industry.
Methods: Making Sense of the Research Journey
Embarking on this exploration of AI in the fashion industry, this research unfolds like a three-stage journey. Let’s navigate through each stage:
| Stage | Content | Method |
|---|---|---|
| Requirement analysis and system design | – Compare the garment development process between existing AI-based design tools and human designers based on the literature review– Identify problems Suggest an AI-based garment development system | Case study |
| System development and Implementation | Development of AI-aided design process– Data Collection for Training (Yoox, Net-A-Porte, Vogue, Tagwalk)– Algorithm design– Training of an image generation model based on StyleGAN2– Front-end Development | Fashion image generation model development based on StyleGAN2– iteration: 200,000 (= 4 epochs)parameters: 28.27 M |
| Pilot test | – Survey on image and system quality satisfaction | Survey |
Requirement Analysis and System Design: Charting the Course
- Setting the Scene: We initiated this research with a case study, examining real-world applications of AI-based fashion design tools. Since garment design is pivotal, we compared these cases with the tried-and-true methods of human designers, seeking commonalities and differences.
- The Exploration: Nine AI-based garment design tools with a commercial track record were scrutinized. Companies like ETRI, LG, Google, Amazon, and OpenAI were among those selected for closer inspection.
- Navigating Complexity: Our expert researchers delved into the data, decoding the intricacies of each AI-based tool. A comparative analysis was conducted, aligning these tools with the established five-step garment development process employed by human designers.
Data Collection: Harvesting Insights
- Scouring the Landscape: The hunt for insights took us to articles and papers, spanning the digital realms of Google, Naver, and Google Scholar. With a focus on the years since 2018, we cast our net wide to gather relevant information.
- Refining the Harvest: From the treasure trove of information, 13 cases were identified, excluding duplicates. The spotlight shone on tools with a proven history of commercial use, eliminating those still in the developmental cocoon.
Coding and Data Analysis: Deciphering Patterns
- In the Laboratory: Researchers with a wealth of expertise in fashion carefully examined each case. A meticulous coding process ensued, dissecting the design processes of AI-based tools. Where opinions diverged, additional searches were conducted, ensuring thoroughness.
- Cracking the Code: Aligning with the human designer’s playbook, commonalities and disparities in the design processes of AI tools were unveiled. These insights formed the bedrock for proposing a new system.
System Proposal: Blueprinting the Future
- The Art of Design: Riding on the revelations from case studies, a new AI-based design system took shape. A collaborative effort involving fashion and computer engineering researchers spanned six months. The aim was to craft a system seamlessly integrating with the human designer’s creative process.
- The Blueprint: This envisioned system boasted four modules, aligning with a user-centered (designer-centered) workflow. The ultimate goal: a tool that augments human design processes rather than overshadowing them.
System Development and Implementation: Bringing Ideas to Life
- Choosing the Cornerstone: The StyleGAN2 algorithm emerged as the linchpin for our AI-based garment development system. Renowned for diversity and image quality, it formed the basis for our technological endeavor.
- Training the Model: The system’s neural network was fed a diet of 52,000 images from 168 leading fashion brands. These images, sourced from global platforms like Yoox, Net-A-Porte, Vogue, and Tagwalk, were the building blocks for training the model. This model, in turn, could generate new fashion images with style variations.
Pilot Test: The Real-World Expedition
- Putting it to the Test: The developed system wasn’t confined to the drawing board. A pilot test ensued, involving a diverse group of eight designers in South Korea. These designers, ranging from SMEs to large corporations, engaged with the system for ten days in December 2022.
- Gauging the Response: Post-test interviews and surveys were conducted, assessing factors like service quality perception, design outcome evaluation, and the intent for continuous usage. A detailed feedback mechanism, including a Likert scale, captured the nuanced responses of the participants.
In essence, our research journey was a meticulous expedition, navigating through the complexities of AI in fashion, decoding tools, proposing new systems, and finally, putting theories to the test in the real-world terrain of designer creativity.
Results and Discussion
Case Study Summary (Table 2): In our exploration of nine AI-based garment design tools, we juxtaposed their functionalities with the intricate stages of human designers’ development process. The results, distilled in Table 2, spotlight where these tools align or diverge from the conventional human-centric design journey.
Results and Discussion on “AI in Fashion Design“
Case Study Summary (Table 2): In our exploration of nine AI-based garment design tools, we juxtaposed their functionalities with the intricate stages of human designers’ development process. The results, distilled in Table 2, spotlight where these tools align or diverge from the conventional human-centric design journey.
| AI design generation model | Human design process | |||||
|---|---|---|---|---|---|---|
| Data analysis | Determination of the season’s concept | Design generation (method) | Design modification | Design finalization | ||
| Internal data (brand internal data) | External data (global fashion trends) | |||||
| AI Fashion Market Platform | X | ◯ | X | ◯ (GAN) | X | X |
| style AI | X | ◯ | X | ◯ (GAN) | ◯ (GAN) | X |
| Tilda | X | X | ◯ | ◯ (GAN) | X | X |
| Project Muse | ◯ | X | X | ◯ (Trained neural network) | X | X |
| Lab 126 | X | X | X | ◯ (GAN) | ◯ (GAN) | X |
| Coded Couture | ◯ | X | X | ◯ N/A (not applicable) | X | X |
| Hybrid Design | ◯ | ◯ | X | ◯ (GAN) | X | X |
| 8 by Yoox | ◯ | ◯ | X | ◯ ( N/A) | X | X |
| Dall-E | X | X | X | ◯ (Diffusion model) | ◯ | X |
- Data Research: Merging Internal and External Insights
- The Human Touch: Human designers extensively delve into internal brand data, a stage present in four out of nine cases. This includes analyzing customer profiles for personalized recommendations.
- AI’s Inclusion: Some tools extend the reach to external data, incorporating global fashion week data and social media trends. However, limitations arise as trends are provided without considering brand identity or seasonal concepts.
- Concept Formation: The Missing Muse
- Human Designers’ Approach: Crafting design concepts is integral for human designers. However, only one case, Tilda, reflects this stage, generating inspiration images based on a given design theme.
- AI Limitation: Most AI tools lack this stage, potentially resulting in less brand-specific ideation and a reliance on technology over design development knowledge.
- Design Generation and Modification: The Canvas of Creativity
- Image Composition Focus: Half of the cases center on GAN-based image composition and text-to-image technologies. However, some tools merely convert text to images without creatively generating designs.
- Human Touch: Only three cases permit modifications after AI-generated design, highlighting the crucial role of human designers in refining creative outputs.
- Finalization Process: The Artistic Full Stop
- Not Universal: Not all cases include the finalization process. This means that the AI tools may fall short in fully realizing the end stages of the design process, leaving room for human intervention.
In Summary: The main drawback of existing AI-based garment design tools is the challenge of faithfully reflecting the nuanced intentions of human designers. While these tools excel in trend analysis and image generation, they often lack the holistic perspective of the human touch. The call emerges for an AI tool that seamlessly integrates fashion domain knowledge for a more nuanced and authentic design output.
Suggestion of AI-Aided Design Process: Bridging the Gap
Understanding the Complexity: Garment design is a nuanced, cyclical process with multiple thinking methods applied at each stage. Our analysis reveals that current AI tools fall short of covering the comprehensive human design process.
Proposed AI-Based Garment Development System (Fig. 1): In response to this gap, we propose an AI-based garment development system that harmonizes with the human design process. The system, comprising four modules, integrates the five stages of the garment design process. These modules act as bridges between the prowess of AI technology and the intricate world of human creativity.
In essence, our proposition aims to elevate AI in fashion design beyond trend analysis, incorporating a more comprehensive understanding of the fashion domain. By integrating AI into the creative workflow of human designers, we envision a symbiotic relationship that enhances efficiency without compromising the unique touch of human intuition and expertise.
Fig. 1: A Blueprint for Integration – This schematic illustration lays out the four modules of our proposed AI-based garment development system, seamlessly embedding AI into the iterative stages of the human design process.
As we delve into the next section on system development and implementation, the practicalities of bringing this vision to life unfold. The journey continues as we transition from conceptualization to tangible innovation, striving to bridge the gap between the precision of AI and the artistry of fashion design.
AI-Aided Design Process
Module 1: Building a Database of Company’s Internal Stuff
This part is about gathering and studying data, both from inside the company and from the world outside. The first module creates a dataset using the company’s own data, and the second one grabs info from the big world. Then, there’s a third module that acts like a storage space where users keep useful words and pictures during the whole process. The fourth module actually designs the clothes and lets users tweak them. Because designing clothes is a bit like doing a dance – you can shuffle the steps around depending on what you need.
Module 2: Global Fashion Trends
The second module looks at the cool trends in fashion from big runway shows. It grabs data automatically from places like TAGWALK or Vogue’s US website and makes a database. This database helps keep track of what’s hot in the fashion world, and the info is used in the designing process.
Module 3: The Design Storage Space
Module 3 is like a storage room for all the cool words and pictures users pick up in Module 1. Users can organize it however they like – by season, type of clothes, or whatever works for them. It’s handy for finding inspiration and creating a mood for designing clothes.
Module 4: Designing and Creating with AI Magic
The last module is where the real magic happens. Users can upload or grab images from their storage in Module 3 and make new clothes. They can also change things up – like colors, shapes, patterns – until they’re happy. It’s a bit like playing with Lego blocks but for fashion. The AI helps by generating lots of different designs. Users can keep what they like, change it up a bit more, and even share it with others to get opinions.
To make all this work, they’re using a fancy AI model called StyleGAN2. It’s like a really smart artist that can make pictures of clothes. The more you use it, the better it gets at understanding what you like. So, it’s a bit like having a super creative buddy who helps you make awesome designs!
Pilot Test of Fancy AI Clothes Making System
The smart folks made a program to let fashion designers check how well their AI system works. They looked at both the numbers (like how fast it works) and what designers thought about it. The numbers were good, and the designers were pretty happy with the AI-made clothes. So, it seems like the AI did a great job and even exceeded what the designers expected.
Conclusion Time
Designing clothes is a big process, like building a whole story. You need to think about what’s trendy worldwide and what fits your brand. If you skip even one step, making clothes that people love and recognize as your brand can get tricky. Most AI systems for fashion only focus on trends and making pictures, missing out on the middle steps. This study tried to fix that by comparing how humans and AI design clothes. The study says it’s a big deal because it brings together computer smarts and fashion sense to make a system that really works.
What’s Special About This Study?
This study did more than just talking about what’s already out there. It brought together tech experts and fashion pros to make a new AI system and actually put it to the test. It’s like making sure that the cool tech stuff fits well into how real designers work. The study says it’s a big deal because it’s not just looking at cases but actually creating something useful.
What’s Next?
The study admits it’s not perfect. They used a fancy algorithm called Style GAN2 but didn’t compare it to other ways of making images. They also mention that the AI system they made has limits, like it works best for dresses and skirts. They suggest looking into these things for future studies. So, there’s more work to be done, but they’re on the right track.