# Getty v. Stability AI Case

Getty v. Stability AI Case refers to separate US and UK lawsuits by Getty Images against Stability AI, alleging unauthorized use of copyrighted photos to train Stable Diffusion, raising legal questions about AI training data and fair use.

The Getty v. Stability AI Case encompasses two parallel legal proceedings filed by Getty Images against Stability AI, the developer of the Stable Diffusion image generation model. The disputes center on allegations that Stability AI copied millions of copyrighted photographs from Getty's databases to train its artificial intelligence system without authorization. The cases, brought in the United States and the United Kingdom, have become landmark examples of the tension between [generative-ai](https://www.wikiprompt.org/wiki/generative-ai) development and existing copyright law.

Getty Images, a major stock photo agency, initiated legal action in both jurisdictions in early 2023. The company argued that Stability AI's use of its proprietary images for training [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) models constituted copyright infringement on a massive scale. Stability AI, in response, maintained that its practices fell under legal exceptions for text and data mining, and in the US context, the doctrine of fair use. The outcomes of these cases are closely watched by technology companies, content creators, and legal scholars, as they could set precedents for how [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) models are trained in the future.

## Background: Stable Diffusion and Training Data

Stable Diffusion, released in August 2022 by Stability AI, is a [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) model capable of generating detailed images from text prompts. The model was trained on a dataset called LAION-5B, which contained billions of image-text pairs scraped from the internet. Getty Images alleged that a significant portion of this dataset included its copyrighted photographs, many of which were watermarked or contained embedded metadata identifying them as Getty content.

The training process for such models involves feeding vast numbers of images into a [neural-network](https://www.wikiprompt.org/wiki/neural-network) architecture, typically a [u-net](https://www.wikiprompt.org/wiki/u-net) combined with a text encoder. During training, the model learns statistical patterns linking visual features to descriptive words. Getty argued that this process necessarily involved reproducing and storing copies of the original images, at least temporarily, and that the resulting model could generate outputs that closely resembled specific copyrighted works.

## US Proceedings: Getty Images (US), Inc. v. Stability AI, Inc.

In February 2023, Getty Images filed a lawsuit in the United States District Court for the District of Delaware. The complaint alleged direct and indirect copyright infringement, violation of the Digital Millennium Copyright Act (DMCA) for the removal of copyright management information, and trademark infringement related to the generation of images bearing Getty watermarks.

Getty claimed that Stability AI had copied over 12 million photographs from its databases without permission. The company sought statutory damages, which could amount to up to $150,000 per infringed work, potentially totaling billions of dollars. Stability AI moved to dismiss several claims, arguing that the DMCA claim was flawed because the removal of metadata occurred during the training process, not in the distribution of the model itself.

In August 2023, Judge William Bryson partially granted Stability AI's motion to dismiss, allowing the copyright infringement claims to proceed but dismissing the DMCA claim and the trademark claim. The court found that Getty had plausibly alleged direct infringement, as the training process involved copying images. However, the DMCA claim was dismissed because Getty failed to show that Stability AI distributed the images to the public after removing metadata. The trademark claim was dismissed because the generated images were not likely to cause confusion as to the source of the photographs.

The case continued through discovery, with both parties exchanging evidence about the composition of the training dataset and the technical details of the model. In late 2024, the court denied summary judgment motions from both sides, allowing the case to proceed to trial. A trial date was set for late 2025, though settlement negotiations were reported to be ongoing.

## UK Proceedings: Getty Images v. Stability AI Ltd.

In January 2023, Getty Images filed a separate lawsuit in the High Court of Justice of England and Wales against Stability AI Ltd., the UK-based subsidiary of the company. The UK claim focused on copyright infringement, arguing that Stability AI had copied and stored Getty images in the UK as part of its training operations.

The UK case raised distinct legal questions because of differences in copyright law. In the UK, there is no general fair use doctrine; instead, there are specific exceptions, including one for text and data mining for non-commercial research. Stability AI argued that its activities fell within this exception, but Getty countered that the training was commercial in nature and that the exception did not apply to the creation of a commercial product.

In July 2023, the High Court allowed Getty's claim to proceed, rejecting Stability AI's application to strike out parts of the case. The court noted that the issues raised were substantial and required full consideration. The case was stayed pending the outcome of the US proceedings, as the parties sought to avoid duplicative litigation. However, in early 2025, the stay was lifted, and the UK case resumed, with a case management conference scheduled for mid-2025.

## Legal Arguments and Defenses

Stability AI's primary defense in both jurisdictions was that training an AI model on copyrighted images did not constitute infringement because the model did not reproduce the images in a recognizable form. The company argued that the training process involved transforming images into mathematical weights and that the final model was a functional tool, not a copy of any particular work.

In the US, Stability AI invoked the fair use doctrine, citing factors such as the transformative nature of the use, the fact that the images were used for non-expressive purposes, and the lack of market harm. The company pointed to previous cases involving search engines and book digitization, where courts had found similar uses to be fair. Getty, however, argued that the scale of copying and the commercial nature of the use weighed against fair use, and that the model could generate images that competed with Getty's own products.

In the UK, the defense centered on the text and data mining exception under the Copyright, Designs and Patents Act 1988. Stability AI argued that its training was a form of computational analysis of data, which was permitted for non-commercial research. Getty countered that Stability AI was a commercial entity and that the exception was not intended to cover the training of commercial AI products.

## Implications for the AI Industry

The cases have significant implications for the broader [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) ecosystem. Many AI companies, including [openai](https://www.wikiprompt.org/wiki/openai), [anthropic](https://www.wikiprompt.org/wiki/anthropic), and [google-deepmind](https://www.wikiprompt.org/wiki/google-deepmind), rely on large-scale web scraping to build training datasets. A ruling against Stability AI could force these companies to license training data or alter their practices, potentially increasing costs and slowing innovation.

Conversely, a ruling in favor of Stability AI could establish a broad right to use copyrighted materials for AI training, which content creators and publishers have opposed. The cases are part of a wave of litigation, including lawsuits by authors against [openai](https://www.wikiprompt.org/wiki/openai) and by visual artists against other AI companies, that seeks to define the boundaries of permissible use.

The outcomes could also affect the development of [large-language-model](https://www.wikiprompt.org/wiki/large-language-model)s and other generative systems. If training on copyrighted data is restricted, companies may need to rely on public domain or explicitly licensed data, which could limit the diversity and quality of AI outputs. Some companies have already begun entering into licensing agreements with content providers, such as [openai](https://www.wikiprompt.org/wiki/openai)'s deals with news organizations, as a proactive measure.

## Technical and Ethical Dimensions

The cases also highlight technical and ethical questions about how AI models are trained. The [neural-network](https://www.wikiprompt.org/wiki/neural-network) architecture used in Stable Diffusion, which includes a [u-net](https://www.wikiprompt.org/wiki/u-net) and a text encoder, is designed to learn the statistical distribution of images. Whether this process constitutes "copying" in a legal sense is a matter of ongoing debate.

Ethically, the cases raise concerns about the rights of individual photographers and artists whose work is used without compensation. Getty Images has positioned itself as a defender of creators' rights, while Stability AI has argued that its technology enables new forms of creativity and that overly restrictive copyright laws would hinder progress.

Some researchers have proposed technical solutions, such as [data-augmentation](https://www.wikiprompt.org/wiki/data-augmentation) techniques or the use of [model-pruning](https://www.wikiprompt.org/wiki/model-pruning) to reduce reliance on copyrighted data. Others have called for the development of training datasets that are fully licensed or composed of public domain works. These approaches remain experimental, and the legal landscape will likely shape their adoption.

## Current Status and Future Outlook

As of mid-2025, both cases remain unresolved. The US case is scheduled for trial in late 2025, while the UK case is in its early stages. Legal experts expect that the US trial will be closely watched, as it could produce the first major judicial decision on AI training and copyright in the United States.

Settlement remains a possibility, as both parties have incentives to avoid the uncertainty of a trial. However, the stakes are high, and any settlement could set a benchmark for licensing fees or establish a framework for future agreements. The cases are also likely to be appealed, regardless of the initial outcome, meaning that final resolutions may take years.

The Getty v. Stability AI Case is part of a broader global conversation about the regulation of [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence). Governments in the European Union, Japan, and other jurisdictions are considering or implementing rules on AI training data. The decisions in these cases could influence those regulatory efforts and shape the future of the AI industry for years to come.

---
Source: https://www.wikiprompt.org/wiki/getty-v-stability-ai-case
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-12T16:23:24.116866+00:00