When a London court ruled against Shein in its copyright infringement lawsuit against Temu, the fashion industry got a rare, public look at how copyright law performs under real competitive pressure. The ruling matters beyond fast fashion: it signals that the legal scaffolding many brands assume protects their visual output — including imagery produced by generative AI tools — is shakier than it looks. If you commission AI-generated fashion visuals, or if your brand's aesthetic is being scraped to train someone else's model, here is what the law currently covers, what it does not, and what you should be thinking about.
Key takeaways
- Copyright in AI-generated images is contested in most jurisdictions; courts have generally required meaningful human authorship before granting protection.
- The Shein v. Temu UK ruling illustrates that claiming copyright infringement on visual output at scale is legally complex, even when copying appears systematic.
- Training generative models on copyrighted fashion imagery sits in a legal grey zone that regulators in the EU are beginning to address — but have not resolved.
- Brands that want enforceable IP should document the human creative decisions behind any AI-assisted work, not just the prompt.
- The gap between what feels like copying and what courts will protect is wide — and currently widening.
What did the Shein v. Temu ruling actually decide?
In a case that WWD described as centring on what Shein called copyright infringement on "an industrial scale," a London court ruled against Shein in its copyright lawsuit against Temu. Just Style, the GlobalData-owned apparel trade publication, reported the outcome as a clear defeat for the claimant.
The specifics of the court's reasoning matter. Copyright in fashion is not a blanket protection for a style, a colour palette, or a general aesthetic. It attaches to specific, original creative works — a particular graphic print, a precisely drawn illustration, a photograph with a defined composition. When two fast-fashion platforms are both producing thousands of SKUs a week, often drawing from the same trend signals and wholesale suppliers, proving that one copied the other's specific protected expression — rather than both independently arriving at a similar garment — is genuinely difficult.
For brands using generative AI tools, this is the first lesson: the law protects specific works, not vibes. If your AI pipeline produces imagery that looks like a competitor's output, or if a competitor's AI produces imagery that looks like yours, the question a court will ask is not "does it look similar?" but "is there an original, human-authored work being copied?"
Does copyright protect AI-generated fashion images?
This is where the law is most unsettled, and where the gap between creative industry assumption and legal reality is widest.
In most jurisdictions — the UK, the US, and across the EU — copyright has traditionally required a human author. The US Copyright Office has consistently declined to register works produced entirely by AI without meaningful human creative input. UK law has a specific provision for computer-generated works (Section 9(3) of the Copyright, Designs and Patents Act 1988), which attributes authorship to the person who made the necessary arrangements for the work to be created — but courts have not yet tested this provision extensively against modern text-to-image models.
What this means in practice: if you type a prompt into a generative image tool and accept the output unchanged, your claim to copyright in that image is genuinely uncertain. If, on the other hand, you make iterative creative decisions — selecting from multiple outputs, editing, compositing, directing lighting and composition through a series of deliberate choices — you are building a stronger case that a human author is present in the work.
For fashion brands, the practical implication is documentation. Keep records of the creative process: the prompts you refined, the outputs you rejected, the edits you applied, the human art direction that shaped the final image. That paper trail is your best evidence of authorship if the question ever reaches a court.
What about training data — can someone use your images to train their model?
This is the question that is generating the most anxiety among designers and creative directors, and for good reason. Generative image models are trained on large datasets of existing images. If your brand's photography, lookbooks, and campaign imagery are publicly accessible online, they may already be part of a training dataset somewhere.
Whether that constitutes copyright infringement is being litigated in several jurisdictions simultaneously, and no jurisdiction has produced a definitive answer. The arguments on each side are real: training a model on copyrighted images without a licence looks like copying; defenders argue that the model learns statistical patterns rather than reproducing specific works, and that this resembles how a human designer learns from looking at thousands of garments.
The EU AI Act is the most significant regulatory intervention so far. It requires providers of general-purpose AI models to publish summaries of the copyrighted content used in training, and to comply with opt-out mechanisms under EU copyright law. This does not resolve the underlying question of whether training itself is infringing, but it creates a transparency obligation that brands can use — if your imagery was used in training without a licence, you at least have a route to finding out.
The Business of Fashion has tracked the broader tension between creative industry interests and AI developers on this question, and the consensus among IP lawyers covering the sector is consistent: the law is moving, but it has not caught up with the technology.
What is and is not protected in AI-assisted fashion design?
It helps to think in layers.
What is more likely to be protected:
- A specific graphic print or textile design with clear human authorship, even if AI tools assisted in its creation, provided the human creative decisions are documented.
- A photograph or campaign image where a human photographer made compositional, lighting, and editorial choices.
- A collection of works where the selection and arrangement itself reflects human creative judgment.
What is much harder to protect:
- A general aesthetic, a colour story, or a silhouette direction — these are ideas, and copyright does not protect ideas.
- An AI-generated image produced by a simple prompt with no documented human creative process.
- A style that has become common in a market segment, even if your brand pioneered it.
What is currently unresolved:
- Whether training a commercial generative model on your images without a licence is infringement.
- How courts will apply existing authorship requirements to AI-assisted workflows where human and machine decisions are intertwined.
- Whether the EU AI Act's transparency provisions will create enforceable rights for brands whose content was used in training.
What should brands and designers do right now?
You do not need to wait for courts to resolve every open question. There are practical steps that reduce your exposure and strengthen your position.
-
Document human creative decisions. For any AI-assisted image you want to protect, keep a record of the prompts, the iterations, the editorial choices, and any post-generation editing. The more clearly you can show that a human author shaped the work, the stronger your copyright claim.
-
Register your most valuable works. In jurisdictions where registration is available (the US being the clearest example), register original works — including AI-assisted ones where you have strong authorship evidence. Registration creates a public record and, in the US, is required before you can sue for statutory damages.
-
Audit your training data exposure. If you are building or fine-tuning a proprietary generative model, work with legal counsel to ensure your training dataset is either licensed, in the public domain, or covered by a defensible fair use or text-and-data-mining exception in your jurisdiction.
-
Use the EU AI Act's transparency provisions. If you operate in the EU or sell to EU consumers, the Act's requirements on training data disclosure give you a mechanism to investigate whether your imagery was used without consent.
-
Do not rely on style as a moat. The Shein v. Temu case is a reminder that looking similar is not the same as copying a protected work. Your competitive advantage in visual output needs to rest on something courts can actually protect — specific works with clear authorship — not on a general aesthetic that any generative model can approximate.
How does this change the way you should commission AI-generated fashion imagery?
The practical answer is: commission it the way you would commission any creative work, with the same attention to authorship and documentation. Brief your creative team or agency to treat AI tools as a production instrument, not a replacement for human creative direction. The art director who decides the mood, selects the outputs, directs the retouching, and signs off on the final image is the author. Make sure that person's role is visible in your process records.
For brands working with external AI image platforms, check the terms of service carefully. Many platforms claim ownership of outputs, or grant you only a licence rather than full copyright assignment. If you are building a campaign around AI-generated imagery, you want to own the work — or at least have a clear licence that covers commercial use and sublicensing.
FAQ
Can I copyright an image I made with a text-to-image AI tool? In most jurisdictions, a simple prompt-to-output image has weak copyright protection because courts require meaningful human authorship. If you made iterative creative decisions — refining prompts, selecting and editing outputs, directing composition — you have a stronger claim. Document your process.
What did the Shein v. Temu UK case decide about AI and copyright? The case centred on alleged copying of specific designs at scale, not AI specifically. The London court ruled against Shein, illustrating how difficult it is to prove copyright infringement when both parties draw from similar trend sources and the claimed works lack clear, distinctive authorship.
Is it legal to train an AI model on fashion brand images scraped from the web? This is unresolved. Arguments exist on both sides, and courts in multiple jurisdictions are actively considering the question. The EU AI Act introduces transparency requirements for training data, which is a step toward accountability but not a definitive ruling on legality.
Does the EU AI Act protect fashion brands whose images were used in AI training? The Act requires providers of general-purpose AI models to disclose summaries of training data and respect opt-out mechanisms under EU copyright law. It does not definitively rule training infringement, but it gives brands a transparency tool and a compliance framework to invoke.
What is the fastest way to strengthen IP protection for AI-assisted fashion visuals? Document human creative decisions at every stage, register key works where registration is available, review platform terms of service for ownership clauses, and work with IP counsel to build a process that makes human authorship visible and provable.
Does copyright protect a fashion brand's overall aesthetic or visual style? No. Copyright protects specific original works, not styles, aesthetics, colour palettes, or silhouettes. Those are ideas, and copyright law does not protect ideas — only the specific expression of them in a particular work.
Further reading
- Shein Loses UK Copyright Lawsuit Against Temu — WWD
- Shein loses UK copyright battle against rival Temu — Just Style