From Algorithm to Runway – Redefining Fashion Design Through AI

Imagine you can not only design personalised garments within minutes, but also try on these self-designed creations virtually. Imagine you can explore different styles that would normally require immense imagination and time, all while receiving styling tips based on your fashion preferences and past wearing experiences. This futuristic form of fashion is already in the making with algorithmically generated fashion, also known as AI fashion.

While not without its challenges, the integration of artificial intelligence in fashion design can change the fashion ecosystem. By bringing innovative and new layers to fashion design, it can take fashion design beyond the limitations of human cognition and express identity in novel and unique ways.

OpenAI’s artificial intelligence chatbot ChatGPT garnered immense popularity at the beginning of this year, causing a sudden rise of interest in artificial intelligence. ChatGPT is able to produce human-like responses at a level that makes it difficult to distinguish from a real human response. This sparked controversy and worry, but it also shone a light on the development of artificial intelligence in various fields, and how AI has the possibility to change these fields completely. The fashion industry, from design to marketing and production, is no exception to the radical changes led by artificial intelligence programmes such as ChatGPT.1

In terms of fashion design, I focus my attention on programmes such as Midjourney, an AI programme that is used mainly to generate images. By providing text- and image-based prompts using the “imagine” command on the social platform Discord, users can generate visually stunning artworks in mere minutes. In a way, these images are the imagined products of the algorithm on the basis of your request. Midjourney is also capable of creating realistic photographic images. It is therefore possible to generate photorealistic images of fashion shoots (see photo above), runways, museum spaces, or garment designs ready to be produced.

Adidas’ digital Ozworld collection2 (see Figure 1) showcases the brand’s foray into AI fashion design as one of many other fashion brands. It is not surprising that (a part of) the fashion industry has adopted artificial intelligence in their processes. After all, fashion and technology have always gone hand in hand. In his book Techno Fashion, Bradley Quinn writes that “fashion itself can be considered as a history of technology.”3 The advancements in technology brought forth by the Industrial Revolutions have always had direct impacts on fashion manufacturing and fashion design; for example, the use of water and steam power resulted in the mechanisation of fashion manufacturing, which only accelerated after switching to electricity. Furthermore, electronics and new information technology made it possible for fashion industries to go online. With recent developments in the field of digital fashion brought forward by the digital technologies of the fourth Industrial Revolution, such as the implementation of AR filters, blockchain technology and the increasing integration of digital fashion in metaverses, it was only a matter of time before algorithms would be thrown into the mix.

Fig. 1. Adidas Ozworld - Avatar pose, from media kit.

AI algorithms are especially appealing to fashion brands because they are capable of analysing large amounts of data in a relatively short time. This means that algorithms can not only identify trends quickly, but they can also generate responses to those trends, making it easier to create designs with a higher chance for ‘success’ on the market. As a result, algorithms can make fashion designers more efficient and productive. Algorithms can also help enhance creativity and innovation if used to create design concepts with which fashion designers can work to create (physical) clothing. Algorithms can inspire designers by providing novel insights and suggestions for their designs. This can push a designer outside of their own boundaries of creativity, sometimes beyond the confines of human cognition. While not always practical or possible in real life, fashion designers can either recreate these algorithmically generated designs digitally or alter their designs to fit within the confinements of the physical fashion ecosystem.

Finally, what makes AI fashion most interesting for fashion design is the improved personalisation and customer experience it brings with it. Through AI-driven recommendation systems, consumers can discover unique style choices that align with their personalities. Algorithms can build forth on past experiences of success and fail and cater to specific preferences and diverse identities. An algorithmically generated image can be taken for what it is, it can be altered by the designer/consumer, or even by the algorithm itself when asked. In other words, including customers in the process of generating AI-fashion results in truly personalised fashion for each customer. If humans work in synergy with AI technologies, they can create designs and, in turn, experiences that are even more personalised than buying ready-to-wear from the shop.

Fig. 2. “Fashion editorial photoshoot, futuristic streetwear outfit, earth tones --ar 3:4”.

Continuing with this idea of personalised fashion creation, algorithms can also break boundaries, such as traditional beauty standards and gender norms. If algorithms are fed diverse visual data, including but not limited to skin colour and body types, they will in turn generate more diverse images. This way, algorithms can make it easier to visualise designs for people with diverse body types and cultural backgrounds, all while keeping in mind one’s style preferences. Fashion designers can more easily break away from outdated or less diverse reference frameworks. In other words, algorithms can bring forth a computer-generated inclusivity that can open up the boundaries in fashion design.

Fig. 3. “3D render, digital fashion full body, fashion magazine shoot featuring a black and green futuristic digital streetwear fashion floating --ar 3:4”.

While the possibilities of AI fashion are enormous for fashion design, I also remain sceptical of its use for several reasons. First of all, there are some ethical considerations of which we must remain critical. The most important one is the role of the traditional (physical) fashion designer, especially if algorithms are normalised and consumers are capable of designing their own clothes. What will happen to the fashion designer? On the one hand, one could argue that this is a positive development, making the fashion industry more accessible to all. On the other, while algorithms are capable of producing fashion design on a technical level, they are unable to create human connections and master human emotions. Thus, to create visually stunning but also emotionally intricate AI-fashion designs, both human and machine are required.

Besides the role of the fashion designer, there are also questions to be asked about bias and privacy. After all, algorithms function because of the data that they are fed. In the case of image-centred algorithms such as Midjourney, that data is visual. In other words, the designs that Midjourney creates are based on what it has already seen visual-wise. That is why it can create designs that resemble, for example, Christian Dior’s well-known little black dress (see Figure 4).

Fig. 4. Four variations of “A dynamic pencil sketch of a black dress in the style of Christian Dior”.

However, these databases are not always ethically sourced. Permission is often not given or not asked at all, and the data is not always diverse (e.g., people with white skin are often more represented than people with darker skin). This leads to biased algorithms with as a result the exclusion of demographic groups or a stereotypical view of these groups. Only if algorithms are trained with diverse source material, and if this material is ethically sourced and transparent, can we expect algorithms to generate inclusive images.

Next to these ethical considerations, we must not forget that fashion is unpredictable: one can never expect what will be in fashion the next season. With AI fashion, however, there is the risk of trend-focused design, or in other words, homogenisation. This is a direct effect of an overabundance of trend-based analysis and design by algorithms. Fashion design in such a way will only accelerate and inflate an already existing loop of fashion design referencing its own references. This could lead to an increased decline in quality, as well as dull and predictable designs.

This is a genuine concern, but instead of stifling creativity, AI can also act as a catalyst for innovation in fashion design. Through the use of AI, colours, styles and patterns can be combined to create novel and unexpected designs that, in turn, can inspire designers again to go beyond the traditional boundaries of fashion design. Furthermore, these ‘new’ designs can be fed to algorithms as new data. As a result, you will have an ‘endless’ loop of new and innovative designs.

All in all, there is no denying that AI is reshaping the fashion industry as we know it now. Even though AI fashion is in its beginning stages, the integration of AI in fashion design is being met with both excitement as well as worry.4 As long as we remain critical of the AI-algorithms that are used by a designer, and the data resourced for these algorithms, I believe that AI fashion can bring new layers to fashion design that include personalisation and increased consumer experience, but that can also break with traditional beauty standards and gender norms. With AI fashion, the relationship between fashion and technology can be explored even more, which begs the question of what direction fashion design will go in the future.

The generated images used in this article were generated using Midjourney v5.1.

Notes

1. https://medium.com/@myschang/unlocking-the-potential-of-personalized-fas...
2. https://news.adidas.com/originals/adidas-originals-introduces-first-of-i...
3. Quinn, Bradley. Techno Fashion. Bloomsbury Publishing, 2002, 3.
4. https://www.forbes.com/sites/theyec/2023/02/21/artificial-intelligence-i...