The impact of AI on IP Law in the world of fashion

Generative AI has the potential to have a significant impact on the world of fashion.

Generative AI is a more creative form of artificial intelligence technology, capable of generating different types of content, including text, imagery, audio, amination and other types of data.

It differs from conventional AI which relies on set procedural rules, which although well suited to pre-defined tasks, has more limited applications in the creative process.

Why is Generative AI attractive to fashion brands and designers?

Generative AI can be commercially attractive to fashion brands and designers as it cuts costs and time to create fashion collections as well as the associated promotional materials.

 It may also increase efficiency with regard to trend and profitability analysis.  For example, online fashion retailer Shein is famed for its use of AI in the analysis of fashion trends and profitability.

However, Shein is the subject of class action lawsuit before the U.S. District Court for the Southern District of New York over its alleged use of AI technology in the creation of its products by a number of independent designers.

Generative AI can also augment, amplify and arguably supplant the human creative process in the world of fashion.

This area is potentially the area of greatest interest for IP lawyers – can AI truly create new designs or is it simply copying old designs?

 Coco Chanel is reported to have said ‘If you want to be original, be ready to be copied’ and the fashion is arguably an industry based on the reimagining of old design trends and aspects.

 This begs the question: is AI really that different to what has come before it?

Essentially, is Generative AI any different to the conventional fashion designer, seeing historic designs and trends and reimagining – or copying –  them?

Maintaining brand identity

Generative AI could be used to maintain the ‘DNA’ of a fashion designer or brand, for example when the original founding named designer passes away.

 Who keeps the original creative flame burning?

When Christian Dior passed away in 1957, 21 year old Yves St Laurent, according to the previous wishes of Christian Dior, was named principal designer at the House of Dior.

However, Yves St Laurent had worked at Dior for a number of years and had already become an integral part of the creative process. What happens when this is not possible?

An active example of this is designer, Norma Kamali, a 79 year old US fashion designer who has engaged AI consultancy Maison Meta to create AI applications which ‘learn’ her design ethos and then can reproduce new designs in that ethos.

For example Maison Meta’s AI platform was given fifteen thousand images of Norma Kamili swimwear ranges and can now produce design products for ‘new’ designs to that ethos within seconds. It can also produce associated advertising images and copy.

This leads on to another attraction of Generative AI for fashion brands: it can be used to produce advertising imagery in minutes, if not seconds, in different ‘locations’ and with different themes.

Customisation, localisation and personalisation

Maison Meta has also worked with fashion brand Coach to represent Coach’s tabby bag instantly in different locations and to reflect how different countries market their products.  vary tone based on local contexts.    

Images can be produced by Generative AI which are simply not possible with conventional photography or are too expensive to reproduce, providing increased possibilities in the world of fashion

Maison Meta has developed the AI Fashion Week, turning designs created by AI applications into real life products. which have been marketed by online platform Revolve. Further, the success of a billboard advertising campaign by Maison Meta for Revolve led to demand for the AI created fashion designs shown in the billboard advertising.

The attractiveness of AI is therefore clear.

Projected success
A McKinsey Report for the Business of Fashion in December 2023 reported that 25% of AI’s potential in fashion will be driven by creative design and product development.
Further, 73% of executives in fashion said Generative AI will be a priority for 2024.
Equity investment for AI focused startups in the first half of 2023 was US$14.1 billion.

The AI ‘input’ problem: training AI and conflicts with IP law
Given the discussion so far, AI seems like the perfect tool for the world of fashion, so why has this not been exploited further? Conventional fashion designs take inspiration from designs which have gone before, they learn and then reinvent. 
So, how does an AI application learn?
AI applications must be ‘trained’ on thousands, if not millions, of images which are then tagged to enable the AI application to create a design based on prompts.
However, if those images are the subject of copyright, it raises the question of whether this constitutes copyright infringement.
For example, AI diffusion models learn by being fed images and breaking them down, or ‘diffusing’ them, and then learning how to put them back together.
Thousands of images of a cat or dog may be fed into the model and diffused until the AI Application understands what a cat or dog looks like.
It can then create its own images of a cat or a dog based on this learning process.  

English High Court case: Getty Images and Stability AI

This interesting case is unlikely to be decided before mid-2025. It  concerns Stability AI’s alleged use of images by Getty Images in training its applications.
On the 1^st December 2023, the English High Court refused to grant summary judgement on behalf of Stability AI in respect of two claims made by Getty Images.
The High Court refused to find that the alleged actions by Stability AI occurred outside the UK and thus beyond the coverage of the Copyright, Designs and Patent Act 1998 (‘CDPA 1988’) and  secondly with regards the allegation that Stability AI was a secondary infringer of Getty Images copyright that the Act only covered physical articles, not intangible items.
Stability AI is also running a ‘pastiche defence’ to copyright infringement claims by Getty Images asserting that in response to prompts by users, the images it produces are a ‘pastiche’ of thousands of images.
In short, the main defence of Generative AI providers are that their use of others’ images is fair use, but the core allegation of the copyright holders is that training involves the copying of images.  
This is possibly a core problem in the use of AI applications in fashion design as the scope for action to be brought against AI created designs is largely unknown, but potentially of significant impact.

Government action or inaction on the ‘input’ problem

The UK government is grappling with the conflict between the IP rights of creatives such as fashion designers and the ability for AI applications to ‘learn’. It is important that the UK is seen as a jurisdiction where both the AI industry and creative sector can prosper.
In many ways this conflict could be unsolvable at least to the satisfaction of both sides of the argument.  

Ultimately the government, whatever political colour that may be, has to make a decision on whether IP laws and in particular copyright law are amended to take account of this potential conflict of interests.  
Last December, in written evidence to House of Lords Communications and Digital Select Committee inquiry into Large Language AI models, OpenAI claimed that although they intend to ‘respect the rights of content creators’ and ‘actively’ to collaborate with them to improve creative opportunities, ‘it would be impossible to train today’s leading AI models without using copyrighted materials’.
In other words, if its models were limited to materials which were outside copyright, then its models would be trained on materials decades old.
To make the analogy to fashion this might imply the styles of the nineteen twenties and thirties might be making a fashion comeback via AI applications of today.
If a Christian Dior or Cristobal Balenciaga had been limited to learning from the designs of the medieval period, what might their designs have looked like?

UK legal developments

In 2022, the UK government indicated that it might provide for a general exemption under copyright law to allow AI applications to learn.  Carve outs have been allowed for private study, why not for learning by AI applications? At present, the data mining exemption under the CDPA 1988 is relatively limited to computational analysis and for the purposes of research for non-commercial purposes.
Copyright claims can easily be brought against the traditional designing process which look at copyrighted materials for inspiration.
One could argue that AI generated designs should be judged by the same stick and ultimately be assessed based on the final outcome, rather than the “feeding” or “learning” process.
In early 2023 the UK government proposed that it would consult with interested parties in formulating an AI code of conduct. However, in early 2024, the UKIPO announced that this consultation failed to level out the tension at play between rights owners and AI developers.
The UK government continues to think on the matter. In response to a question in the House of Lords by Lord Holmes on 9^th May 2024, Viscount Camrose, the UK Minister for Artificial Intelligence and IP stated that his ‘crystal ball prediction’ was that copyright law would change and that the resolution of the ‘input problem’ would take three elements:

1.  some change to copyright law
2. technology would be used to achieve a resolution
3.  international standards would be agreed.

Viscount Camrose accepted that the laws regarding ‘machine generated content’ were ‘fairly elderly’. We will come back to this point on the ownership of AI generated works later.

The AI ‘output’ problem

It would appear on the face of it that existing IP laws in the UK can deal with any ‘output’ problem with regards infringement of IP rights and copyright.
Unlike the imput side there does not appear to be any need to amend existing laws to deal with this side of the AI equation. 
If an AI application were to produce a fashion design which was a copy of, or took a substantial part, of a copyrighted work then it would appear that this would clearly be copyright infringement which could be actionable under existing laws.  
However, there does appear to be an issue with the ownership of AI created fashion works and whether such works can be the subject of IP rights.

The AI ‘creation and protection’ problem for IP law
UK copyright and design law does provide for the creation of works which are computer generated. The author is deemed to be the person by whom the arrangements necessary for the creation of the original work are made.  
However, at least under copyright law, the work must be original, resulting from the author’s free and creative choices.
AI creative fashion designs may fall into a range of creative works and each would need to be assessed on their own merits.
Yes, computer generated AI fashion works could be protected under UK copyright law but equally they might not depending on the originality of the work.
This might suggest that these provisions, which were created long before the rise of today’s AI, may not be fit for purpose.

In the US recent decisions, most notably those of the Review Board of the United States Copyright Office on 5^th September 2023 with regards the work ‘Théâtre D'opéra Spatial’  and the decision of the 18^th August 2023 of the US District Court for the District of Columbia with regards the work ‘a recent entrance to paradise’ copyright for AI works have been denied on the basis of the lack of a lack of human authorship and in the words of the District Court a lack of the ‘guiding human hand’.

However, this contrasts to the position of China where Beijing Internet Court in November 2023 held that images of a young woman created by Stable Diffusion by the Plaintiff was protectable under Chinese copyright law. The images had been posted on social media platform Xiaohungshu using the hashtags #AI, #AIillustration and others and ultimately were copied by the defendant.

The AI ‘liability’ problem for IP law

Who is liable when an AI application infringes IP rights, for example in the creation of a fashion design?
Some have argued that AI applications be given legal personality like a human being or a company.
This idea might not be as surprising as one might think- for example, a company has legal personality and is not a human being. A similar model could apply to AI.
However, if an AI application were to be given legal personality, there is the potential of rights.  
One could also look to Roman Slave Law, under Roman Slave Law the owner of a slave, here arguably the AI application, would be held liable for the crimes of the slave.
Nevertheless in the world of the internet it might be difficult to identify the “slave owner”, and one might have a ‘I am Spartacus’ problem. The Google France decision with regards keyword advertising however could be used to give guidance on the liability issues relating to AI.

Generative AI user liability
Could users of Generative AI tools be held liable for infringing works created by the Application?
The user will likely have little knowledge of how the AI application was trained.
When the user is the ‘creative’, it may be easier to check whether the created works are potentially infringing others IP rights.
However, with an AI application the user will largely be in the dark.
Such works are created by prompts, and some commentators have argued that AI applications will have to build in ‘guardrails’ to avoid IP infringements as much as possible.
Further, there is the so-called ‘Snoopy problem’ put forward by Professor Matthew Sag of Amory University School of Law.
The more a likeness is protected by copyright works, for example the cartoon character dog Snoopy, the more likely an AI application is likely to copy it as a representation of a typical dog.
This problem is sometimes referred to as ‘the Italian plumber’ problem with reference to Mario from the computer game Super  Mario Brothers. In fashion we might have the ‘red sole shoes problem’ in connection with Christian Louboutin, we suppose.

Generative AI and online search

Generative AI also has the potential to change the way one searches for fashion products.

The online search market has been relatively stable for twenty years, and the way one searches for branded products has not fundamentally changed either with text searching producing long lists of websites ranked by relevance, influenced by keyword bidding on branded and unbranded terms.

However, Generative AI search heralds the potential for conversational searching which will fundamentally change the way online search operates and looks. Further, chatbots such as ChatGPT could change the way keyword advertising works.

There are presently issues with Generative AI search tools. In essence, if they do not know the answer they make things up. A lack of accuracy is holding Generative AI search tools back.

Generative AI search tools may never be perfect - and might never have to be. Trade mark law accepts that the average human consumer is imperfect.

Potential challenges for Generative search AI- “Fuzzy matching searches” might lead a chatbot being confused between NIKE or NIKF.

- Chatbots may have issues with indirect confusion and the free riding of reputable trade marks. For example, personal assistants such as Alexa have a tendency to recommend the market leader.

- Can a chatbot be trusted not to recommend counterfeit products?

- How can legislation make it clear when advertising or influencers are utilising Generative AI?

So where are we with Generative AI and fashion?
The potential for Generative AI for the economy is huge, and Generative AI applications are particularly attractive for the fashion market. There are clear commercial drivers for fashion brands and fashion designers to use Generative AI applications to save costs and to speed up and augment the creative process in the world of fashion.

Will Generative AI replace the human fashion designer? Probably not, but the application of Generative AI in the world of fashion does raise particular issues for IP law, mostly notably with regards how AI applications learn and whether works created by AI applications can be protected under AI law.