Photographers Found Out from a Press Release
How Getty and Shutterstock each handled the licensing of their libraries for AI model training — Shutterstock's contributor fund arrangement and Getty's lawsuit against Stability AI — told from the position of the photographers whose frames were in the files.
When AI Licensing Became Someone Else's Decision
In July 2023, Shutterstock announced that it had signed a deal with OpenAI to license images from its library for use in AI model training. The announcement was made through a press release. Contributors — the photographers and illustrators whose work comprises that library — learned about it the same way everyone else did.
Shutterstock had been in a licensing relationship with OpenAI since 2021, providing images used in the development of DALL·E. The 2023 agreement extended and expanded that arrangement. In parallel, Shutterstock launched what it called a Contributor Fund: a pool of money distributed among contributors whose images were determined to have been included in training datasets. The fund was not opt-in in any meaningful sense. Images already licensed to Shutterstock under standard contributor agreements were subject to the terms of those agreements, and those terms had always granted Shutterstock broad rights to sublicense content for commercial purposes. The question of whether "AI training" fell inside that language was answered, in Shutterstock's reading, by the company itself.
The contributor fund distributed payments, though the methodology for calculating individual shares was not made fully transparent to contributors. The underlying licensing framework meant that a photographer who had uploaded work years before large language models and diffusion models existed was now a participant — however passive — in the commercial transaction Shutterstock had negotiated. The agreement was announced; the terms were the terms.
Getty's Litigation and What It Did and Did Not Change
Getty Images took a different position. Rather than licensing to AI developers, Getty filed suit against Stability AI in February 2023, alleging that Stability AI had scraped and used millions of images from Getty's archive without authorization or compensation to develop the Stable Diffusion model. The suit was filed in the United States District Court for the District of Delaware. A parallel action was brought in the United Kingdom. Getty's complaint pointed to images bearing its watermark appearing in Stable Diffusion outputs as evidence that its visual archive had been used in training.
The lawsuit positioned Getty as the photographers' defender — the agency standing between its contributors and unlicensed exploitation of their work. That framing was not entirely unfair. Getty did not license Stability AI. Getty's contributors did not receive a portion of an AI training deal, because no such deal existed on Getty's side. In the narrow sense, Getty's action preserved a boundary that Shutterstock had already chosen to dissolve.
But the structural reality for photographers was not dramatically different between the two agencies. In neither case did photographers negotiate. In neither case were they party to the decision. At Shutterstock, the decision was to license and construct a fund. At Getty, the decision was to litigate. Both were institutional choices made by companies about assets those companies held sublicensing rights to. The photographers found out from the press release.
The distinction that mattered was downstream: if Getty's litigation succeeded, it would establish that training on scraped commercial image libraries without authorization constituted infringement. If it settled, the terms of any settlement would likely remain confidential. Either way, the remedy — if any — would accrue to Getty Images as an institution, not automatically to individual contributors as a right.
The Agreement Architecture Underneath
Stock photography agreements are not written for a world in which a licensee's product is itself a content-generating system. When a magazine licensed a Getty photograph in 2008, the use was bounded: a specific image, a specific placement, a defined run. The contribution agreement a photographer signed granted the agency rights to sublicense that image for defined or broadly defined purposes. "Commercial use" has long been the relevant category, and it has always been capacious.
The US Copyright Office began formal study of AI and copyright in 2023, including questions about whether the ingestion of copyrighted works for model training constitutes fair use — a question the courts have not yet settled definitively as of mid-2025. The outcome of that analysis will shape how future licensing disputes are adjudicated. It will not retroactively restructure agreements photographers signed before the question was live.
Shutterstock's contributor fund therefore represents one model for the post-hoc accommodation of contributors whose work enters AI training pipelines through an agency's sublicensing decision. The fund exists. The amounts distributed are not individually disclosed. Whether the aggregate pool is proportionate to the commercial value generated for Shutterstock by the OpenAI relationship is a calculation no contributor can independently make.
Getty's litigation represents a bet that the licensing value of its archive can be defended through the courts rather than transacted away. A favorable outcome would mean that Stability AI — and, by implication, other developers who trained on scraped commercial archives — would owe licensing fees, damages, or both. That would be commercially significant for Getty. Whether photographers would share meaningfully in any recovery depends on contributor agreement terms, on how damages are calculated, and on choices Getty would make about distribution.
What the Sequence Established
The two paths — licensing with a fund, and litigation — have created a rough template for how large stock agencies can respond when AI developers want their archives. A third path, taken by some independent archives and individual photographers, is proactive refusal, with rights assertions and metadata designed to signal non-consent. Whether that signal is honored depends entirely on whether the downstream developer chooses to honor it, and on what legal architecture eventually governs training data.
What did not happen, at either agency, was a collective negotiation in which photographers set the terms. The frame is in the library. The library is a commercial asset the agency manages. When that asset becomes newly valuable in a way no one anticipated when the contribution agreement was signed, the agency decides what to do with it. Photographers found out from a press release. They found out what the decision was, not that a decision was being made.
