AI and images

Do picture agencies accept AI-generated images? Policies compared

A person stands at the open cabin door of a white high-wing aircraft.
Boarding a light aircraft.Photograph by Robert Harper. Robert Harper Photograph Library, Archive.London. From the archive, shown for illustration.

Most of the large picture libraries refuse AI-generated submissions, one welcomes them with a label, and the reasons given are nearly all about paperwork. Read together, the policies describe what buyers are paying for.

Read the agencies' AI policies side by side and they turn out to be one policy: we sell pictures whose origin we can account for.

Mostly, no. Of the five large libraries whose stated positions we checked on 18 November 2026, four refuse AI-generated submissions from contributors and one accepts them under a label. The more interesting finding, and the angle here, is in the reasons. Almost none of them is aesthetic. The agencies do not say generated pictures are ugly or unwanted. They say they cannot establish who owns them, who should be paid, or what the picture is a picture of. Read as a set, the policies are an accidental definition of the product: an image with an accountable origin.

This article is general information, not financial or legal advice. Policies change, so check the linked page before relying on any row. Checked on 18 November 2026.

Which picture agencies accept AI-generated images?

Organisation Accepts AI-generated submissions? Stated position Source and date
Getty Images (including iStock) No Stopped accepting AI-generated images in September 2022; a year later told contributors that it "does not accept files created using AI generative models", including Adobe's Firefly tools PetaPixel reports, 21 September 2022 and 21 September 2023
Shutterstock No "Shutterstock will not allow AI-generated content to be submitted by contributors for licensing on our platform" Contributor help article, 16 July 2025
Adobe Stock Yes, labelled Must be flagged "Created using generative AI tools"; no artist names, real people or fictional characters in prompts, titles or keywords Contributor guidelines, updated 11 June 2026
Alamy No "We do not accept any AI-generated imagery on the Alamy platform"; classed as unsuitable material and removed Alamy statement, 13 February 2025
Pond5 No "At this time, Pond5 does not allow the submission of AI-generated content in our Marketplace" Help centre article, 5 November 2024
World Press Photo (contest) No Synthetic images and generative fill disqualify an entry; AI enlarging tools are a breach Rules page for the 2027 contest

Two notes on the table. Getty's row rests on trade press reports of its notices to contributors; Getty's own policy pages are not covered here, so check them for the current wording. And the table covers submissions. Several of these companies sell generated images made with their own tools, which is a different matter and is dealt with below.

The wire services are not covered here. The Associated Press publishes standards on generative AI, and we would sooner leave a row out than fill it from memory.

Why do most libraries refuse AI-generated submissions?

Shutterstock gives two reasons, and they are worth quoting in substance because they are so unromantic. First, contributors must be able to prove that they own the intellectual property in what they submit, and because AI models draw on the work of many artists, ownership of the output cannot be assigned to one person. Second, the company wants artists paid when their work is used in training, and where the source model is unknown it cannot confirm that everyone who contributed is compensated.

Pond5's two reasons are the same two: no individual can be credited as author, and it cannot ensure that every artist whose work fed the output is paid.

Getty's chief executive, Craig Peters, put it in a sentence when the ban was first reported: "There are real concerns with respect to the copyright of outputs from these models."

A picture library's whole business is a warranty. When a publisher licenses an image, the fee buys less the pixels than the library's promise that it has the right to grant the licence and that nobody will come knocking afterwards. We describe that function in what a picture library is. A submission whose authorship and inputs cannot be stated is one the library cannot warrant, however good it looks.

How does Adobe Stock's approach differ?

Adobe takes the opposite commercial bet with similar caution. It accepts generative AI content, and its guidelines invite contributors to use the tools to make work that fills "content needs within the collection". The conditions are the notable part:

  • Every such file must be flagged as created using generative AI tools, and a second box ticked where people or property are fictional.
  • Prompts, titles and keywords must not contain artist names, real people or fictional characters, nor references to copyrighted works or wording that implies a real newsworthy event.
  • Content that depicts or is based on an identifiable person needs a model release, including where a real person's photograph was used as a prompt.
  • Breaches may lead to removal of content or closure of the account.

So even the library that says yes has built a regime to control the three risks the others cite: confusion with photographs, imitation of named creators, and the likeness of real people. Its rule on artist names connects to the wider question of whether style is protected.

The label is also where trouble shows. Adobe's community forum carries threads from contributors who say genuine photographs were rejected for a missing AI flag. A rule that depends on telling the two kinds of image apart inherits the difficulty described in how to tell whether a photograph is AI-generated.

Where is the line between AI editing and AI generation?

Every policy has to answer this, because AI is now inside ordinary editing software.

Organisation Allowed Not allowed
Alamy Retouching-style tools such as denoise on original photographs Generative Fill, or anything that substantially changes an image's reality
Adobe Stock Extending a background, removing distractions, recolouring a background, without the AI label Adding or changing a primary subject without the AI label
Getty Images (as reported in 2023) Standard editing in Photoshop and Illustrator; 3D renders Files created with generative models, including Firefly tools
World Press Photo Limited use of tools such as denoise and automatic adjustments, if nothing is added or removed Generative fill; AI enlarging such as Adobe Super Resolution and Topaz Photo AI; smartphone HDR, portrait and panorama modes

The common thread is information. Tools that work with what the camera recorded are tolerated. Tools that add what it did not are either banned or made to wear a label. World Press Photo is the strictest because its subject is news: it bars AI enlargers on the ground that they "introduce new information". We apply the same test to archives in AI colourising and upscaling of archive photographs.

Alamy is candid about enforcement. Its statement accepts that AI content is not always labelled and can be hard to spot, and says uploads are continually monitored. A ban is a contract term before it is a filter.

Do the same companies sell AI-generated images themselves?

Some do, and there is no contradiction once the reasoning above is understood. Shutterstock's policy notes that its own image generator was trained on licensed Shutterstock data. Getty launched a generator in September 2023 trained on its own creative library. In both cases the company knows what went in and has an arrangement with the people who made it. The objection was never to generation. It was to generation with unknown inputs.

The distinction between creative and editorial material matters here too. On Getty's fourth-quarter 2025 earnings call, held on 16 March 2026, its chief executive said the company does not license editorial content for AI training. Editorial photographs are records of real events and real people, and the companies that hold them are treating that as a separate category of asset.

What does this mean for the value of authentic photographs?

We will not put a number on it, and nobody should take pricing guidance from a table of contributor rules. Three observations seem safe.

First, the policies create two shelves. On one sits material whose origin is documented: who made it, when, with what, under what rights. On the other sits everything else. Buyers who need a warranty, which includes most publishers and advertisers, shop from the first.

Second, the first shelf is harder to stock than it used to be. A library must now satisfy itself that a submission is a photograph, and the surest evidence is an original: a raw file, a negative, a transparency, in sequence with its neighbours.

Third, labelling is becoming infrastructure. Adobe's flag, the metadata field described in IPTC Digital Source Type, and signed credentials all exist to keep the shelves apart once files leave the library. Anyone can inspect what a given file declares with a metadata viewer or a Content Credentials checker.

None of this predicts what any picture will fetch. It describes what the largest intermediaries have decided they can stand behind.

Where a film archive fits

The Robert Harper Photograph Library is the work of one photographer over about thirty-five years: 95,487 unique photographs by its inventory count, of which 52,684 are analogue, held as film and prints, and 42,803 are digital, including Nikon NEF raw files. It is owned by 825 Ventures Ltd, the owner and rights holder. More than half of the archive is still on film and has yet to be scanned, and the counts have not yet been independently reconciled.

That is an account of origin, not a statement that any image is ready for any use. People, brands and products appear in these photographs, and what a licence covers is set out in our image licensing terms and decided case by case.

Agencies, publishers and picture researchers who want material with a physical original behind every file can find the practical details on the page for picture agencies.

Frequently asked questions

Do stock agencies accept AI-generated images?

Most large ones do not accept them from contributors. On the pages we checked on 18 November 2026, Shutterstock, Alamy and Pond5 refuse AI-generated submissions, and Getty Images has been reported as refusing them since September 2022. Adobe Stock is the main exception: it accepts generative AI content that meets its rules and is labelled.

Why do Getty and Shutterstock refuse AI-generated submissions?

The reasons they give concern ownership and payment, not taste. Shutterstock says contributors must be able to prove they own what they submit and that it cannot confirm every artist behind a model's output is compensated. Getty's chief executive cited concerns about the copyright of outputs when the ban was reported in 2022.

Can I sell AI-generated images on Adobe Stock?

Yes, subject to its guidelines, last updated on 11 June 2026. Content must be flagged as created using generative AI tools, must not use artist names, real people or fictional characters in prompts, titles or keywords, and needs a model release where it depicts or is based on an identifiable person. Review standards still apply.

Do photography competitions allow AI images?

The major photojournalism contest does not. World Press Photo's rules for 2027 require entries in its main categories to be made with a camera, disqualify synthetic images and generative fill, and treat AI enlarging tools as a breach. Limited AI-assisted adjustments such as denoise may be allowed if they add and remove nothing.

Is editing a photograph with AI tools the same as an AI-generated image?

The policies draw the line at generation. Alamy accepts retouching-style tools such as denoise on original photographs but may fail images altered with Generative Fill. Adobe Stock requires its AI label when a tool adds or changes a primary subject, and not for extending a background or removing a distraction.

Sources

Published by Archive.London. Corrections are welcome under our editorial policy.

Licensing the archive for AI and data projects

95,487 photographs from one photographer and one rights holder. Read the dataset card, search the 503-folder inventory, discuss a sample set under NDA, or arrange a first look as scanning proceeds.

See AI and data licensingWrite to the archive