Forum

Licensed to Show, Not to Train: The Line Getty Just Drew

A bright circular ring light glowing against a dark background

On 21 June, Getty Images announced a multi-year partnership with OpenAI that brings Getty’s licensed content into search and discovery experiences in ChatGPT. The deal covers a library of more than 400 million assets, including editorial coverage of sport, entertainment and news. Getty was explicit about one limit: the agreement is for display only and does not cover training AI models.

A few days earlier, on 15 June, a magistrate judge in the studios’ copyright case against Midjourney in Los Angeles refused the company’s request for broad information about how Disney, Universal and Warner Bros. use AI internally. Midjourney had argued that the studios’ own AI practices were relevant to its defense. The judge found them irrelevant to whether Midjourney had infringed, and the studios agreed to provide information only about their consumer-facing AI applications.

Two separate events, one shared lesson: rights in the AI era are no longer one bundle. They come in layers, and each layer is being negotiated, or litigated, on its own.

Display is not training

For years, a license to use an image mostly meant a license to reproduce and display it in a defined context. Generative AI added a new use that older contracts never contemplated: feeding images into a model so that it learns from them.

Getty’s announcement makes the distinction concrete. ChatGPT users may see Getty pictures, credited and licensed, when they search for news or events. That does not give OpenAI the right to train on those pictures. A single relationship between a rights holder and an AI company can now include some uses and exclude others in plain terms.

For photographers and illustrators who license their work through agencies, this is a useful precedent. It shows that “AI use” is not one thing, and that contracts can and should separate display, search, training and generation.

What everyone else does is not a defense

The Midjourney ruling addresses a different argument that often comes up in public debate: that companies complaining about AI training use AI themselves. Whatever the merits of that point as commentary, the judge’s decision suggests that, in an infringement case, what the plaintiffs do with their own tools is largely beside the point. The question is what the defendant did, and whether that conduct infringed the plaintiffs’ rights.

That is worth remembering for creators too. Using AI in your own practice does not weaken your rights over your own work, and it does not license you to use anyone else’s.

Reading your own contracts

Many creative professionals have agreements with agencies, platforms, publishers and clients written before generative AI existed. This is a good moment to check them.

  • Does the agreement say anything about AI training, or only about reproduction and display?
  • Can the agency license your work to AI companies, and on what terms?
  • Are you paid separately when your work is licensed for training?
  • When you deliver work to a client, does the contract say whether they may use it to train models?

Where the answers are unclear, clarity is worth negotiating. The Getty deal shows that sophisticated parties are now writing these distinctions into their agreements.

What it means for competitions

Award programs license entrants’ work too, usually for promotion. The International AI Award rules say that submitted images are used to promote the award and its activities, with credit given, and that entrants must own or have permission for what they submit. As AI uses multiply, clear and limited terms of this kind, stating what an organizer will and will not do with entries, become part of what makes a competition trustworthy.

The takeaway

The era of all-or-nothing arguments about AI and creative work is giving way to something more detailed and more practical. Display, search, training and generation are different uses with different values. Creators who understand the difference, and write it into their agreements, will be in a far stronger position than those who assume a license is a license.