Can a fashion designer use GenAI to continue their legacy?

News that a renowned designer is training an AI to create in her style prompts serious questions over IP ownership, say Magdalena Borucka, Louise Popple and Margarita Taliadoros of Taylor Wessing.

Fashion designer Norma Kamali has revealed that she is training an AI model to create designs in her own unique style—allowing her to continue her legacy when she is no longer designing. But what are the legal implications of using generative AI tools to continue the legacy of designers after retirement, or even death, in the UK?

Do AI-generated works qualify for IP protection?

The law of designs is by far the most important to the fashion sector. Protection can be in the form of a registered design (which lasts for a maximum of 25 years), an unregistered design (called a supplementary unregistered design which lasts for 3 years) or another form of unregistered design called 'design right' (which lasts for 10 years from first marketing or 15 years from creation where there is no marketing). Supplementary unregistered designs are most commonly relied upon.

Design legislation largely recognises 'computer-generated works' (ie works with no human author), meaning that AI-created fashion designs should be protectable under the law of designs assuming they meet the usual requirements in the legislation.

However, the legislation relating to supplementary unregistered designs is silent about computer-generated works. Although there are good arguments that this form of protection should also be available for computer-generated designs, given the uncertainty, designers might want to consider applying to register any key computer-generated designs.

Copyright is less relevant to the fashion sector, largely because garments must qualify as "works of artistic craftsmanship" as well as being original for copyright to subsist. The position as regards computer-generated works is more complicated under the law of copyright.

It is not clear whether AI-generated works satisfy the current UK/EU test for originality or whether human input to the creative process is also required.

Although UK copyright law specifically recognises computer-generated works, the interplay between this provision and the originality requirement has not been considered by the courts. To mitigate this risk, designers should ensure they have some input into the design process (eg, by refining the work).

Trademarks (eg for colours and logos) are of obvious importance in the fashion sector but are not considered here.

Who owns the IP in AI-created works?

A related issue is who owns any designs or copyright (if it satisfies the originality test) in AI-created fashion works. Where the legislation specifically recognises computer-generated works (see above), it also provides that the person who made the arrangements necessary for the creation of the work will be considered the author (and therefore usually the first owner) of any resulting copyright/designs in the work.

The generally held view is that this means that it is not possible for the AI tool itself to be the author and therefore owner of any resulting designs or copyright in an AI-generated work (which would reflect the position for patents, which require a human inventor).

While this all seems relatively straightforward, working out who is the person who made the arrangements necessary for the creation of the work is unlikely to be easy, as acknowledged by the UK IP Office in a 2021 consultation. If interpreted broadly, it could include the supplier of the AI system itself even if they did not input any data/undertake any specific training.

Having said this, those relying on commercial AI tools, like OpenAI, should own all IP rights in their outputs since suppliers of commercial systems largely assign all IP rights to their users as standard. Nevertheless, terms of service should be checked/negotiated and written agreements covering IP ownership and use should be entered into with anyone involved in the AI-generation process. This should also ensure that the right person owns any supplementary unregistered designs in any AI-generated outputs (the legislation being silent on ownership).

Risks of IP infringement with AI-generated designs

Infringement of both forms of unregistered design protection and copyright all require copying. When using AI to create fashion items, it is possible that copying occurs when the AI tool is trained/"learns" or when it creates outputs.

What if Kamali only inputs her own designs to train the AI tool? The answer is that, at the training/learning stages, there should not be any infringement by Kamali. However, it is still possible that there could be infringement at the output stage eg if the AI tool was pre-trained by its supplier by crawling the internet (as is usually the case), and any resulting output generated copies any third party design-/copyright- protected works (in a way that meets the relevant test for infringement).

This risk could be mitigated by using a bespoke AI tool that has not been trained in this way, by conducting internet searches against outputs, and by negotiating appropriate terms of service with the AI supplier.

What if Kamali uses third-party works to train the AI eg to train it in evolving trends? The position would then be more complex. Fashion designs and trends are intrinsically transient - one season might see high-waist jeans, white and chic aesthetic, the other season might see low-waist baggy jeans, red and sporty aesthetic.

AI tools deprived of information available in the public domain on trends and consumer hype would likely generate collections that are 'out of fashion'. However, the use of works of other designers (without permission) to train AI tools comes with obvious risks of copyright and design infringement (at the 'training' stage and possibly also at the 'learning and 'output' stages, subject to the relevant provisions in the legislation being met). This risk could be mitigated by using written words to describe trends/train the AI as opposed to images/other data.

The position is similar when it comes to registered designs, although infringement does not require copying so arguably there is even greater scope for infringement in the above scenarios. Again, this is assuming the other conditions for infringement in the legislation—such as commercial use—are met.

The other issue to bear in mind is the risk that any AI-generated 'output' creates a false message leading the public to believe that the design was endorsed, recommended or approved by a third party when it was not. This can constitute passing off. This should make designers wary about designing fashion items, whether AI-generated or not, that look like they come from a third party (without permission).

This means that care must be taken when training and using AI tools. The permission of any third-party IP owners should preferably be obtained before the tool is trained on their works and—either way—care should be taken to ensure works generated do not infringe any third-party IP rights or suggest a connection with a third party when that is not the case.

What does this mean for designers using genAI to continue their legacies?

Generative AI has the potential to revolutionise the fashion industry. The launch of AI Fashion Week in 2023 and the showcasing of AI-generated pieces in London Fashion Week in February 2024 are milestones in the integration of generative AI in the sector. However, the IP position will nearly always need to be considered.

More generally, whether generative AI can 'fill the shoes' of influential and talented designers, who set the tone and culture of a brand itself, remains to be seen. However, it might help speed up the design and commercialisation process (with input from humans) and improve the customer experience through better marketing and customer service.

Magdalena Borucka is a senior associate at Taylor Wessing. She can be contacted at m.borucka@taylowessing.com

Margarita Taliadoros is an associate at Taylor Wessing. She can be contacted at m.taliadoros@taylorwessing.com

Louise Popple is a senior counsel – knowledge at Taylor Wessing. She can be contacted at l.popple@taylorwessing.com