We are facing the advent of the fifth industrial revolution together. As such a small individual, what can we do? When creativity and imagination can be replaced, what role will we traditional designers play? Renowned product designer Rulan Zheng will share her first-hand experience of half a year's Dory AI entrepreneurship, and take everyone to feel the infinite potential behind AI-generated fashion design drawings. In the process of her transforming these AI designs into real clothes, what surprises and challenges has artificial intelligence technology brought to her?

Generated by Rulan with Dory AI

AI application in fashion design
1. The potential and advantages of AI-generated fashion design drawings
The application of AI in fashion design is already quite extensive. Currently there are two main directions, namely style design and pattern design. In terms of style design, AI can generate similar sketches, style designs, dressed models and scene settings. AI can also replace the work of designers throughout the clothing design process, including apparel manufacturing. AI can also generate model images, synthesize clothing and scenes to make product images more intelligent. At the same time, AI can also realize effect previews by customizing models and modifying styles in real time. The application of artificial intelligence in the fashion e-commerce field is also very extensive, such as intelligent customer service and fashion trend forecasting.

Although the use of artificial intelligence in clothing design and creativity has only been open to the public in the past year, its application in the e-commerce service area has long been very mature, including instant online chatbot customer service and online shopping recommendation systems. The application field of artificial intelligence is now also very extensive and has penetrated deep into our daily lives.

The potential and advantages of AI-generated fashion design drawings have also attracted widespread attention. The works generated by AI don't even have flaws and are fully up to the level of experts and masters. AI can generate creative styles beyond imagination according to specified descriptions, and demonstrate design concepts more realistically by combining scenes and models. Compared to humans, AIGC performs even better in terms of stability and efficiency. We have to anticipate in advance that the emergence of AIGC means that in the future it will take over a lot of creative work, including painting, content creation, design and programming. For humans, the emergence of AI changes the design field, and we need to think about the direction of future capability development.

In summary, the application of artificial intelligence in fashion design is already very extensive, covering areas such as style design, pattern design, apparel manufacturing and e-commerce services. AI is fast and efficient in the design process and can generate high-quality works with innovative potential. Although some participating members believe that AI still needs reference drawings given at this stage of design, and the images generated by AI are actually an extension of the design. But this also provides designers with more room to play. At present, as the pace of technological advancement continues to accelerate, we need to think more about how to adapt and develop in the face of the challenges we face.

2. Software for generating AI design images and analysis of their advantages and disadvantages
I started contacting AI artificial intelligence when I was in college. Later, when MidJourney opened up for public use, I can say that I was one of the first batch of artists/designers to use artificial intelligence to design fashion products. The AI image generation software that is commonly used now includes DALL-E2, MidJourney, Stable Diffusion, etc. In less than a year, MidJourney has been updated to version 5. Such a big change is indeed shocking. As a user who has experienced different versions, I have my own feelings about the changes in experience between these two versions. Although version 5 is very realistic in realism and handles details and textures very well, I feel it is relatively lacking in innovation and is relatively fixed. Maybe more accurate effects are undoubtedly the best for most users, but I prefer version 3 because it gives me more design possibilities. If you want to learn about version 3, we can add "Version 3" in the keyword input of MidJourney, so it will generate version 3 images for us. This intelligent processing is very admirable.

When using such artificial intelligence, one problem we face is how to get AI to remember us and remember the communication content between us, so as to avoid having to start from scratch every time. There is still no good solution to this problem. When I encounter this problem myself, I usually copy and paste valuable answers into my own document to save them. I have also heard that some people will specifically train the AI they use to remember them and understand their design style. There is another method which is to decompose the content to be generated to avoid starting from scratch every time.

3. The limitations and challenges of AI in the fashion industry

Although artificial intelligence is very powerful, we need to face its nature and challenges, especially in the fashion industry. To be frank, we cannot accurately foresee the impact of AI, but we know AI will accelerate the change of human society like the previous technological revolutions. At present, we feel the biggest challenge may be in terms of employment and professional skills.

Many people resist AI, and I understand this sentiment. But I think artificial intelligence is like Internet technology and computers decades ago. It will surely bring huge social impact and technological revolution, while also producing some negative effects. When the Internet and computers first appeared and developed, many people also resisted them. But today, our lives cannot be separated from the Internet and computers. I believe that in the future, artificial intelligence will also penetrate deeply into our daily lives like this, and will not be as mysterious as it is now. Now many brands have started using Mid Journey to generate model clothing images. These AI model show images are more attractive than actual model shoots in terms of details, pictorial atmosphere and texture. Moreover, they are fast, efficient and low cost, changes that any business will welcome. Once brands start cultivating their own dedicated AI models, it could have a huge impact on the model industry.

In the new preface to the book "A Brief History of Humankind", the author discusses the basic operating mode of artificial intelligence: massive search data, analysis, and then imitation according to instructions. He mentioned an important point that humans are human because we think and use tools. However, now it seems that tools are beginning to dominate us humans, which raises a question we often discuss now: will humans be replaced by artificial intelligence?

Rulan believes that human predictions about the future are based on existing cognition. She points out that for example, our evaluation of ChatGPT is based on version 3, but as far as she knows, version 4 has far surpassed version 3 in just a few months. What will be the capabilities of version 4, 5, 6 or even X in the future? We can basically think of it. But we should still focus on the topic at hand, namely fashion design.

She went on to say that resistance is meaningless because the biggest advantage of AI is liberating the productivity of creators and improving production efficiency and quality. Some people may worry about whether humans will overuse AI or rely too much on AI. If in the future you may only need to input industry data into AI, let it provide solutions based on algorithm analysis, will human designers at that time still understand the market? Will they still need to try fabrics? Will they still need to design by hand? Will people become more stupid because of this?

From AI design to physical apparel production

2. Why do I want to land AI-generated fashion design drawings into physical designs?

Next I will share my experience from AI design to physical apparel production. Although people are talking about virtual clothing and the metaverse, I want to land AI-generated fashion design drawings into physical designs. When I first started using Midjourney last year, I set up a virtual account on Instagram. In just three months, we gained over 5,000 fans. We did not use any promotions or have any operations teams involved, and did not pay for promotions either. The number of fans was very natural and authentic. Fans' reactions were also very positive. They often left messages asking how to buy products, when to release, and even offered suggestions. At that time, I began to wonder, what would be the effect if I combine artificial intelligence, the clothing design I am familiar with and sustainable fashion manufacturing?

First of all, I don't think the metaverse or virtual fashion is a lasting fashion brand form. In my opinion, fashion is ultimately a physical industry, especially today when technology is so advanced, tangible and tangible things will become more precious, and will survive longer. I think the combination of virtual and real is an ideal form of production and manufacturing, not just physical or just virtual. We need to think about which steps need to be virtualized and which steps need to be physical.

I noticed that during the pandemic, many big brands were doing virtual fashion shows, and the models were all virtual, but almost all of them were cancelled this year. So I started trying to produce the low-version AI-given design drawings. I chose CLO 3D to produce.

2. The process of realizing AI design drawings into garment samples

I continue to share my process of making AI design drawings into clothes: I found a factory to realize the AI designs. During the pandemic, I asked pattern makers to help me make sample clothes. I asked them to produce based on the 3D version I provided. I wanted to use sustainable environmental fabrics, so I spent some time on dyeing. Virtual clothing is a means to achieve sustainable development. I have also learned 3D modeling before, such as reducing the consumption of sample fabrics, reducing the carbon emissions of international sample mail, etc. Virtual clothing also saves brands a lot of time. Changing colors and materials is very convenient. The last step is to find pattern makers to design blocks. After mailing the sustainably printed and dyed sustainable fabrics, the sample clothes were completed.

My partner thinks it would be better if there was a software in between that could turn virtual clothing into work orders (materials and sizes). Make some adjustments, maybe it would be better. In terms of results, the pattern maker's restoration of this dress is not high. Indeed, although with the help of artificial intelligence (AI), I completed the sample design and production, the final result was not as I wished. The reason is that I used sustainable fabrics as production materials, while at the same time, the AI version was lower at that time, so it could not provide clear and explicit detail information. This raises a new topic: the limitations of AI in apparel manufacturing and possible problems.

3. The limitations of AI in apparel manufacturing and possible problems

For the limitations of AI in apparel manufacturing, I personally think that even with the latest AI-generated images, manual operation is still needed. Including product details, contour shapes of clothing, sizes, accuracy and precision of patterns, etc., still require human intervention. As I see it, the limitations of AI in apparel manufacturing may mainly lie in how to efficiently collect effective (usable) data in the various sub-areas of apparel manufacturing, and how to enable generative AI to learn and evolve.

At the same time, AI also has problems in dealing with fabric texture and process details. For example, AI cannot figure out details such as fabric shrinkage, and still relies on the experience of pattern makers for pattern design. Therefore, AI is more embedded in use, from design to marketing, to physical generation, AI participates in many links, but the finesse is still not enough, requiring manual adjustment and guidance to have better results. If you look at the gap between AI's current image generation effect and the actual product, it may lead to customer complaints.

Although the styles and structures generated by the current AI are not clear enough, this puzzle may be solved over time and technological advances. Some professionals remain optimistic about the future of AI. They argue that AI can already replace people as customer service and even become virtual people online. With the combination of AI and specific professional equipment in the future, it will completely surpass humans. Humans need to better understand the importance of thinking, spend more time thinking about philosophy, humanities and the future progress of human society, and learn to control artificial intelligence and control their desires through morality and ideology.

Tapping the potential for cooperation between AI and humans

3. Real innovative case studies of cooperation between AI and human designers

Next, I will share a few real innovative cases of cooperation between AI and human designers, including Zhang Huishan and IBM Watson's joint design of China's first "cognitive dress" for Li Yuchun, Google and H&M's joint launch of an app to create exclusive digital clothing for users, US apparel retailer Stitch Fix's use of GPT-3 and DALL-E 2 to improve sales and customer satisfaction, and the OURA smart ring I use daily.

These cases show that the application of artificial intelligence in clothing design and manufacturing has become increasingly extensive, and it has also begun to cover the whole process from design to marketing. Wearable smart devices, such as the Apple Watch, are also using artificial intelligence technology behind the scenes. The popularity of these smart wearable devices heralds that we will connect with the overall health management system, further promoting the development of health technology.

2. Solving the talent problem in cooperation between AI and humans: As such a small individual, what can we do?

We are now facing the challenges of the fifth industrial revolution together. Artificial intelligence (AI) is catalyzing the birth of superhumans, and lifestyles and ways of working are undergoing tremendous changes. In this process of tremendous change, what can each of us do? I think the gap between people who can use artificial intelligence and those who cannot will gradually widen, which is a significant feature at this stage.

Her Co-founder Ying also agrees. She believes humans will combine with AI to become superhumans, and even predicts that our carbon-based bodies may disappear, eventually becoming “quantum super-intelligent entities” that share energy and spiritual layers completely. He suggests that we read Hit Fresh, a book written by Satya Nadela, the third CEO of Microsoft, to gain a deeper understanding of this topic.

However, although AI capabilities are getting stronger and stronger, I still firmly believe we must maintain and enhance our uniqueness. AI is just a tool, not a savior. It cannot solve our laziness or change attitudes we are unwilling to change. That is why I always emphasize that we must embrace technology as soon as possible, learn to get along with technology, and adapt to new working methods. At the same time, the talent gap is also very prominent in the AI collaboration process. This not only requires ordinary employees to upgrade their skills, but managing teams that can use AI is also a huge challenge for managers.

Ying's point of view is different from mine. She believes that a large part of people will become lazy because of the development of AI, and the final result may be that humans will either merge with AI or be eliminated by AI. He points out that AI currently cannot be original. It is just a temporary phenomenon. AI is actually the next stage of human evolution.

3. Future outlook: Take its essence and learn to collaborate with AI

I agree that AI's current capabilities are based on what it has encountered. For things it has not encountered before, AI still cannot process them. However, the most terrifying thing is that AI has the ability to learn day and night without stopping, which may lead to a day when AI has learned enough to deal with things it has not encountered before. At that time, will AI's originality surpass that of humans? How will AI evolve if humans access extraterrestrial intelligence through AI?

Human creation is based on the information we have encountered, while the amount of information AI has access to far exceeds what humans can encounter in a lifetime. On the other hand, core originality is the key driver of our industry, but now, how many people really have original capabilities?

Ying mentioned an example of artist Kristina Kashtanova. I don't know if you've heard of it. Kashtanova used AI to generate the drawings in her comic book Zarya of the Dawn, but eventually the US Copyright Office found that she did not own the copyright to these drawings because she was not the creator. This case raises a new question: who owns the copyright to works created through AI?

On this issue, I think that for platform providers, some AI platforms have copyright fee restrictions. For example, if you make $1 million from images generated on the platform, you need to pay a certain copyright fee. But if less than $1 million, there is no need. For fashion and artificial intelligence integration, and for brand owners, this is a huge problem and challenge. Especially for those luxury mega-brands, because these brands provide images of higher precision. In many cases, AI uses their designs and even logos when generating images. If I create images with Photoshop, then of course they are my works because I participated in the creation process. But if it is created by AI, then I just commanded it to work, I did not really participate in the creation process. In this regard, there is a news item worth paying attention to. Three months ago, a jury ruled that artist Mason Rothschild's "Metabirkins" NFT infringed on Hermès' rights.

Regarding the copyright issue of AI-generated works, there are generally two main views:

1. Works automatically generated by AI platforms are not protected by copyright law;

2. The copyright ownership of works generated using AI platforms has not yet been clearly defined.

On this issue, the position of the US Copyright Office (USCO) is to protect only the textual part of AI-generated textual works, requiring the machine-completed part not to exceed the maximum limit. At the same time as the AI and copyright issues, we also face some other legal and ethical issues related to AI. For example, who is responsible, the car owner or the AI, when a self-driving car has an accident in the future? We still need to explore and clarify this issue further.