A High Court in London has started hearing the lawsuit that Getty Images has filed against Stability AI, an AI-based text-to-image generator. Getty accuses Stability AI of unlawfully scraping millions of copyrighted images to train their LLMs (Large Language Models).
Justice Joanna Smith, the British High Court judge hearing the case, described it as “highly complex and technical,” raising “numerous novel issues for the court to consider.” She made these remarks during one of the four hearings held earlier this year, before the trial began on June 9, 2025.
What is Getty’s Allegation Against Stability AI?
Getty Images is a licensing platform for copyrighted multimedia content, including photos, illustrations, vectors, videos, and audio files.
The lawsuit against Stability AI was filed in 2023. In January, Getty Images announced in a press release that it had initiated legal proceedings against Stability AI in the High Court of Justice in London. It claimed that the AI startup had infringed copyrights owned or represented by Getty Images.
“Stability AI unlawfully copied and processed millions of images protected by copyright and the associated metadata owned or represented by Getty Images,” reads Getty’s position.
Getty stated that it has licensed its content to other companies for training artificial intelligence systems while respecting personal and intellectual property rights. However, it claims that Stability AI did not seek such a license, instead pursuing “their stand-alone commercial interests” in creating Stable Diffusion, a deep learning text-to-image model released in 2022.
Getty’s Terms of Use explicitly prohibit “using any data mining, robots, or similar data gathering or extraction methods.”
Additionally, Getty raised new allegations about Stability’s potential to produce child sexual abuse material (CSAM) and apply Getty’s watermark on AI-generated images. Justice Smith advised Getty to plead these allegations separately and clarified that Getty’s reference to “pornography and violent images” does not fall under criminal acts, unlike the creation of CSAM.
What are Major Legal Issues?
Oliver Fairhurst, a copyright lawyer specialising in AI, called this case “an absolute behemoth.” He outlined the following legal issues involved:
Trade mark infringement,
Copyright infringement,
Database rights infringement and
Passing off.
Trademark and copyright infringements are common in such lawsuits. A trademark is any graphical representation distinguishing a person’s goods or services from others in the marketplace, while copyright grants exclusive rights to the creator of an original work.
Database rights infringement is slightly different but falls under the broader umbrella of copyright law. A specific provision under the UK’s Copyright and Rights in Databases Regulations 1997 protects data defined as a “collection of independent works, data, or other materials arranged in a systematic or methodical way.”
Similarly, “passing off” refers to the unauthorised use of another person’s goods, services, or business goodwill, potentially causing confusion or deception in the market and leading to unfair competition.
Getty claims that Stability must now prove it did not violate any copyrights, as the AI-generated output closely resembles the original images. Lindsay Lane KC, arguing for Getty, stated that holding users responsible for AI-generated images would assume an unrealistic level of technical literacy. She also highlighted that users might assume the watermarked images have legitimate permission, thereby reinforcing the trademark violation argument.
How does GenAI Generate ‘New’ images from prompts?
Stability Diffusion uses the dataset called LAION-5B, developed by The Machine Vision & Learning Group at Heidelberg University, led by Germany-based Computer scientist Björn Ommer.
Researchers released LAION-5B, a large-scale open dataset, under the Creative Commons (CC-BY 4.0) license. The dataset consists of 5.85 billion links to images hosted online, paired with short descriptions.
When a user enters a prompt, the AI system checks whether the keyword exists in its dataset and identifies which images match the prompt. The AI then generates images from random noise using a process called diffusion, which involves corrupting the image by adding noise.
In the reverse diffusion (or denoising) process, the AI restores the corrupted image into a recognisable form — the final AI-generated content.
It is important to note that Stability AI trained its model on the LAION-5B dataset but does not store the original images. Instead, the model learns patterns and structural similarities to generate “new” images based on user prompts.
Advertisements
Can third parties use copyrighted content to train large language models (LLMs)?
The legal complexity arises from the fact that Stability AI’s current model may not contain exact replicas of Getty images — if it used them at all. Instead, it generates images with structural similarities. Even so, if copyrighted material was used, even temporarily, it could amount to infringement.
According to Section 17(6) of the UK’s Copyright, Designs and Patents Act 1988, making transient copies “incidental to some other use of the work” can still be a copyright violation.
For instance, in 2008, the UK Supreme Court ruled that “making transient copies of copyright material into Random Access Memory (RAM)” is a restricted act. The case examined the legality of “effective technological measures” used by video game console manufacturers to prevent piracy.
What is Stability AI’s Defence?
In 2023, Emad Mostaque, founder of Stability AI, claimed on X that the company’s training data were “ethically, morally, and legally sourced and used.”
However, during the ongoing trial, Stability did acknowledge that there were some instances where the trademarks appeared on AI-generated images. Hugo Cuddigan KC, arguing for Stability, said the images submitted by Getty were “contrived,” implying they were artificially generated solely for litigation purposes.
Before the trial began on June 9, a spokesperson for Stability stated that the dispute concerns “technological innovation and freedom of ideas.” They argued that enabling artists to use these tools falls under fair use and freedom of expression.
Can Stability AI Evade the Liability?
Stability AI may defend itself by comparing its denoising and data-scraping processes to how humans browse the internet using search engines, which is not considered copyright infringement.
Fairhurst noted that Stability AI could argue it used copyrighted content under fair use, having applied creativity to build something new. It might also claim protection under certain legal exceptions and data mining provisions, such as:
Article 4 of the EU Copyright in the Digital Single Market Directive, if the training occurred in the EU
If the text and data mining had taken place in Germany, under section 44b of the German Copyright Act.
If the data mining was done for research and non-commercial purposes accompanied by sufficient acknowledgement, section 29A of Copyright, Designs and Patents Act 1988 in the UK,
Since Stable Diffusion was developed in partnership with a German university, Stability AI could argue the training was for non-commercial academic research. However, post-Brexit, the UK no longer falls under EU jurisdiction, so Article 4 does not apply. Therefore, the provisions of Section 29A of Copyright, Designs and Patents Act may provide Stability with limited respite.
The AI company may also claim the rightful use of data under Section 28A and 29A of Copyright, Designs and Patents Act 198 in the UK, which allows the making of copies and data mining, but only if there is no “independent economic significance” and only if used for “non-commercial research.” Stability AI released its Stabled Diffusion model under the MIT license, a permissive open-source license.
The company may still invoke Sections 28A and 29A of the UK Act, which allow limited copying and data mining, but only when the activity lacks “independent economic significance” and is strictly non-commercial. While Stability AI released Stable Diffusion under the permissive MIT license, it also monetises tools like DreamStudio, blending open-source development with commercial services.
Getty argues that Stability AI built commercial products on the back of content acquired — allegedly — without proper licensing.
Also Read:
Support our journalism:
