Last month Getty Images, one of the largest image libraries in the world, announced a licensing deal with OpenAI. This might seem surprising at first glance: image libraries sell their content through licences and subscriptions, while AI tools generate images for free. Indeed, last year Getty brought a lawsuit against another AI developer, Stability AI, accusing Stability of training its image-generation tool, Stable Diffusion, on millions of Getty’s images without permission (a case I wrote about here). In November, the High Court in London rejected most of Getty Images’ claims, though the decision is currently under appeal.
At the heart of Getty’s complaint against Stability AI is that the developer’s use of Getty’s images – the core of its value – was without consent. Licensed (which is to say, authorised) use of content, on the other hand, is what underpins Getty’s business. This is exactly what the new deal with OpenAI is all about.
While the exact terms of the collaboration are not public, it is described in Getty’s press release as a ‘display’ agreement, whereby licensed content will now appear in ChatGPT’s search results. Crucially, the deal does not mean that OpenAI can use Getty content to train any of its AI tools. This ‘display-only’ model is similar to a recent deal Getty made with another developer, Perplexity, and has significant benefits for Getty. Not only does it provide a new licensing revenue stream, but since it does not extend to AI training, it also protects Getty’s content from being used to improve the capabilities of potential competitors.
Both Getty and the AI developers recognise that, in a world where it is in short supply, authorised, attributed and trusted content carries unquantifiable value. In the context of this particular deal, it makes ‘AI‑powered search and discovery more useful and more trustworthy’, according to Craig Peters, CEO of Getty Images. While access to trusted and ‘authentic’ content is good for everyone, the direct effect of this ‘display’-only type of deal is felt more by the companies involved than by end users of their products. Users of ChatGPT whose prompts return a Getty image will need to get a licence directly from Getty should they wish to reproduce it – as they have always done.
Deals between content creators and AI developers have proliferated over the past couple of years. To date they have mostly involved press publishers, such as Associated Press and Reuters, licensing their content to AI companies for training purposes. But there are increasing numbers of image licensing examples too. Shutterstock has been especially active in seeking partnerships with tech giants such as Nvidia, Meta and Amazon as well as OpenAI. (A merger between Getty Images and Shutterstock was announced last year but abandoned earlier this month.) Getty has struck other deals with interesting collaborative arrangements, specifically by partnering with tech companies to create its own generative AI tools. Getty describes its ‘AI Generator’, based on Nvidia architecture and launched in 2023, as ‘commercially safe’, being trained only on its own licensed data. In so doing, it seeks to differentiate its offering from other tools developed using web-scraped content. As concerns over the legal and ethical risks associated with web scraping intensify, its proposition becomes more compelling.
Licensing deals between creative content providers and AI developers are generally happening at the top end of the market, largely to the benefit of multi-million-dollar businesses. For smaller businesses and startups it is quite a different picture. Concerns about how or even whether work they own has been used in AI development continue to vex creators.
Surveys recently conducted by the UK copyright collecting society DACS reported that 96 per cent of visual artists had not given permission for their work to be used for AI training and nearly two-thirds did not know whether their work had been used in this way at all. Given that the median yearly income for visual artists in 2024 was just £12,500, the opportunity for fair compensation through licensing could not be more vital.
Creators are certainly not taking this challenge lying down. There are more than 100 cases against AI developers in the United States alone, including a class action by a group of visual artists against Stability AI. Artists in the UK are also making their voices heard. It was in no small part the concerted lobbying by musicians, artists and publishers that prompted a recent change in UK government copyright policy away from a pro-developer stance. In Europe, the EU AI Act of 2024 has set the rules that are steadily shifting how participants in this space must operate. Of particular note are the obligations the EU now imposes on developers to be transparent about how their AI tools are trained. This is crucial for creators, since they can hope to secure control and fair compensation for use of their work only if they can first establish where and how it has been used. The transparency issue is under consideration by the UK government, whose latest report acknowledges its significance for creators and developers. For the moment, the UK does not have a comparable piece of legislation, but with Andy Burnham having this week announced that the AI minister, Kanishka Narayan, will be given a seat in Cabinet, perhaps progress will accelerate.
While regulators grapple with the new order, those affected will continue to work out what is possible. Deals similar to the ones Getty and others are striking with developers will surely multiply as the benefits of collaborative licensing models become clear. These rewards must extend to individuals as well as to powerful multinationals. Collective management organisations such as the Design & Artists Copyright Society (DACS) can help secure blanket licences with developers for use of their members’ works in AI training. This will ensure that economic benefits are enjoyed by the artists these organisations represent, with fair compensation through licence fees being passed down to the human authors on whom AI tools ultimately depend.
Emily Gould is assistant director of the Institute of Art and Law.
