- Freelance photos usually belong to the photographer; a paper or agency needs a signed transfer or license to include them.
- Getty Images says it pays contributors when their content goes into AI training sets, but gives no AI rate; its general contributor royalty is 15% to 50% of the license fee.
- The image deals AI companies have announced since 2025 are mostly display licenses for AI search, not training sales.
- Reported 2024 figures run from 5 cents to $1 per photo (Photobucket's range in talks) to $1 to $2 per image (a data company's market estimate); none is a contract price.
Who owns the photos in your photo archive?
Start with who pressed the shutter and under what paper. A staff photographer's work belongs to the employer. A freelancer's work belongs to the freelancer unless a signed agreement says otherwise. An agency usually holds a license from its contributors, not the copyright. Those three facts decide what an archive can license for AI training, whether you are a photo agency, a newspaper picture desk or a photographer's estate.
The statute is specific about publications. "Copyright in each separate contribution to a collective work is distinct from copyright in the collective work as a whole, and vests initially in the author of the contribution." Without a written transfer, the publisher "is presumed to have acquired only the privilege of reproducing and distributing the contribution as part of that particular collective work" and its revisions (17 U.S.C. 201(c)). A newspaper's archive of freelance pictures may hold no AI rights at all.
| Rule or source | What it says | What it means for a photo archive | Date checked |
|---|---|---|---|
| 17 U.S.C. 201(b), works made for hire | The employer is the author of work its staff make on the job | Staff photographers' pictures belong to the paper or agency | 20 Nov 2026 |
| 17 U.S.C. 101, commissioned work | Commissioned work is made for hire only in listed cases, with a written agreement signed by both sides | A freelance assignment is not work for hire by default | 20 Nov 2026 |
| 17 U.S.C. 201(c), collective works | Each contribution keeps its own copyright; the publisher is presumed to hold only the right to use it in that publication | Freelance photos in a paper's archive may carry no AI rights | 20 Nov 2026 |
| 17 U.S.C. 204(a), transfers | A transfer of copyright needs a signed writing | Old verbal deals with photographers do not move ownership | 20 Nov 2026 |
| Columbia Journalism Review, 8 Jul 2026 | Freelance photos at almost every major US paper and magazine remain the photographer's property | Expect the photographer, or their estate, to hold the rights | 20 Nov 2026 |
| Getty Images 10-Q, quarter to 30 Sep 2026 | Contributors are paid when their content goes into AI training sets; staff editorial work is owned outright | Agency archives split into owned and contributed content, with different rules | 20 Nov 2026 |
CJR reports that "At almost every major newspaper and magazine in the United States, photographs produced by freelancers remain the sole intellectual property of their creator." It also reports that the Wall Street Journal's new freelance contract "would assign the Journal primary authorship of all photographs produced on assignment through what’s known as a Work for Hire clause", and that "six hundred and fifty freelance photographers who work with the Journal have signed on" to oppose it (CJR). A new contract covers new assignments. Photos made under older terms stay under those terms unless the photographer signs again.
Photo archive rights check: what must you clear beyond copyright?
Owning the copyright is the first test, not the last. Work through these before you count images.
- Model releases. Copyright is not consent from the people in the frame. Stock and advertising images usually carry releases; editorial archives usually do not. Many buyers training image models ask for releases, or for images where people are not the subject. See the model release guide.
- Property releases for private buildings, artworks and branded products when they are the subject of the picture.
- People in news photos. Public figures at public events carry less risk than private people, children or victims. Mark sensitive images and leave them out.
- Pictures in your files that belong to others: handouts, pool pictures, wire images and pictures sent in by readers.
- Captions and metadata: who, where, when and who took it. Captions are written work with their own copyright, and they are much of what makes a news archive useful for training.
- Credits, watermarks and copyright notices. Keep them in your records, because buyers check them against your rights column.
Getty's own lawsuit shows how much captions matter. Its US case against Stability AI concerns the "alleged unauthorized reproduction of approximately 12.0 million in images from Getty Images’ websites, along with the accompanying captions and associated metadata" (Getty Images 10-Q).
Who buys our photo archive for AI training?
The public buyers fall into three groups. The deals announced since 2025 say more about search display than about training.
- AI search and chat companies that display images. Getty Images and Perplexity announced a multi-year license on 31 October 2025, under which the "Agreement gives Perplexity rights to display images from Getty Images across its AI" search tools (Getty Images). On 21 June 2026 Getty announced a display agreement with OpenAI: "Under the partnership Getty Images’ licensed content libraries will appear across OpenAI search and discovery experiences within ChatGPT" (Getty Images). Both are display licenses.
- Image model developers that train. Getty's filing says "Contributors are compensated for any inclusion of their content in AI data training sets and may share in the revenue generated by AI tools and services trained with their content" (Getty Images 10-Q). Training sets are being built from agency content, and libraries that own their images outright are the simplest sellers to these buyers.
- Data companies that gather images to license on. Their rights vary. PetaPixel reported that the chief executive of one AI data company said "she avoids acquiring content from platforms such as Photobucket because the licensing rights are murky" (PetaPixel, 8 April 2024). Curated archives with clear ownership are the opposite case.
Photographers selling new stock rather than an archive have other routes: see Shutterstock data licensing and the Wirestock review. Brand and campaign libraries are covered in brand and product imagery.
How do photo agencies license images for AI training?
Agencies license in two ways. They build AI tools on content they license, or they license collections to AI developers and pay contributors a share. Getty does both. Its filing says contributors "are compensated for any inclusion of their content in AI data training sets", but it gives no rate for that payment. The rate it does give is its general one: "Suppliers who choose to work with us under contract typically receive royalties of 15% to 50% of the total license fee we charge customers". Getty also owns some content outright, "including content produced by our staff photographers for our editorial product" (Getty Images 10-Q).
An agency can license only what its contributor agreements allow. If your older contracts did not mention AI, you have the same gap a newspaper has. There are two ways to close it: new contributor terms that add AI licensing with a revenue share, and an opt-out for contributors who refuse. Whichever you use, keep a record of which contributor signed which version.
A small agency or a picture library can also license its owned collection directly as a dataset: images, captions, keywords and dates, delivered in batches with a manifest that lists every file and its rights status.
Agencies comparing platforms will find Human Native's former marketplace folded into Cloudflare; our Human Native review covers what its site now tells rights holders.
What is our photo archive worth to AI companies?
No AI company publishes a rate card for photos, and none of the reported figures is a contract price. In April 2024 PetaPixel, reporting Reuters, wrote that Photobucket's chief executive "has discussed rates of 5 cents to $1 per photo and over $1 per video with prices varying depending on the buyer and the types of imagery they are seeking" (PetaPixel, 8 April 2024). The same Reuters reporting quoted the head of the AI data firm Defined.ai saying companies "generally pay $1-2 per image" (VentureBeat, reporting Reuters). Photobucket's images came from users under changed terms. A curated archive with clear ownership is a different product, so treat both figures as context, not a price for yours. These are the figures we found; the list is not exhaustive.
The Copyright Office's report on AI training says licensing "may be workable, at least in certain contexts", and names "popular music and stock photography" as content that "can be licensed in relatively high volumes" (Copyright Office report). Volume helps. So does being the only source of a picture.
- Clear ownership of each image, shown in a rights column.
- Captions written by people who were there: names, places, dates and events.
- Scarcity: decades of local news, events no one else covered, and specialist subjects such as science or industry.
- High-resolution scans of negatives and prints, not web-sized copies.
- Consistent metadata across the collection, so a buyer can filter it.
How is a photo archive priced in AI data deals: per image, flat fee or royalty?
Four structures show up in the public record we read. None comes with a published contract price.
| Pricing model | How it works | What the public record shows | Source or date |
|---|---|---|---|
| Per image | A fee for each image delivered | Reported: 5 cents to $1 per photo in Photobucket's talks, and $1 to $2 per image as a data company's market estimate; no closed deal at either rate is public | Reuters reporting, April 2024 |
| Flat fee | One payment for a defined collection, term and use | Announced image deals with AI companies do not publish their fees | Getty and Perplexity, Oct 2025; Getty and OpenAI, Jun 2026 |
| Royalty or revenue share | A share of what the licensor earns from AI licenses or AI tools | Getty says contributors are paid for inclusion in training sets but gives no AI rate; its general royalty is 15% to 50% of the license fee | Getty Images 10-Q, quarter to 30 Sep 2026 |
| Pay on use | A payment each time content is used by the AI system | Offered for sound effects; no photo deal with this structure is public | Read 20 Nov 2026 |
Per-image pricing rewards volume, which suits large stock libraries. A flat fee suits a curated collection where each image is rare. A royalty keeps you in the upside but needs reporting you can audit. For an editorial archive, a flat fee for a non-exclusive license with a fixed term is the simplest start, and you can ask for a per-image floor so the fee grows if the buyer takes more.
What have courts and the Copyright Office said about photos and AI?
In the UK, Getty Images sued Stability AI, but the court never ruled on whether training infringes. The judgment of 4 November 2025, [[2025] EWHC 2863 (Ch)](https://www.judiciary.uk/judgments/getty-images-v-stability-ai/), records Getty's acceptance that "there is no evidence that the training and development of Stable Diffusion took place in the United Kingdom", so that claim was abandoned at trial (the judgment). What the court decided was narrower: secondary infringement, where it held the model is not an infringing copy, and trade marks. AP reported that "Artificial intelligence company Stability AI mostly prevailed against Getty Images Tuesday in a British court battle over intellectual property" (AP). Getty's annual report says the court "granted Getty Images’ request to appeal the decision on secondary infringement" (Getty Images 10-K). The Court of Appeal heard that appeal on 11 November 2026 (Pinsent Masons), and we found no appeal judgment as of 20 November 2026. Getty also has a separate case in the United States.
The lesson for an archive owner is not that scraping is safe. Lawsuits are slow and uncertain, which is why AI companies pay for images with clear provenance. The Copyright Office's report says "Direct licensing is most common and most promising with respect to corporate entities with catalogs of high-quality and easily identifiable content" (Copyright Office report). A well-captioned agency or newspaper archive fits that description.
What will a buyer's lawyers ask about your photo archive?
Have answers ready for each of these, by image or by batch.
- A rights column for every image: staff, freelance with a transfer, freelance without one, agency, handout or unknown.
- The contracts behind each category, with dates and versions.
- Which images show identifiable people, minors or private property, and which have releases.
- Where a bought or inherited collection came from, with the signed transfer.
- Caption and keyword fields, and who wrote them.
- Opt-outs or takedown requests from photographers or subjects.
- Whether any images are already licensed to someone else on exclusive terms.
Images marked unknown are usually left out, and that is fine. A smaller archive with a clean rights column is worth more to a buyer than a larger one full of open questions.
How we checked this
We read every source on 20 November 2026: the copyright statute on ownership, works made for hire, collective works and transfers on the Copyright Office's official site; Getty Images' own filings with the SEC (its 10-Q for the quarter to 30 September 2026 and its 10-K for 2025) and its own announcements; the UK High Court's judgment in the Getty case and a law firm's report of the appeal hearing; the Copyright Office's Part 3 report on AI training; and press (AP, Columbia Journalism Review, VentureBeat and PetaPixel reporting Reuters). Some sites refuse scripted fetches, so we saved the text we read.
We could not find a closed AI deal with a published per-image price, and the display licenses do not disclose their fees. Release and publicity rules differ by country, and we did not review each one. This page is general information, not legal advice.
Licensing an editorial or historical photo archive: where to start
Start with the part of the archive you own outright: staff work, collections bought with signed transfers, and freelance work under contracts that granted you the rights. Build the rights column for that part first. A photographer's estate holding one person's life's work is often in the simplest position of all, because one person owned most of it.
Origin Data Partners is an independent data partner that helps business owners value and license their company records: the company pays Origin nothing, and a buyer may pay Origin when a deal completes. We help you license your data. We start from a description of the collection and its rights, not the files themselves, and you sign any license yourself. Not the owner? Archivists and advisers can introduce the collection's owner.
Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.
Common questions
Should we license our photo archive exclusively or non-exclusively?
Non-exclusive is the usual start. It lets you license the same images to several AI companies and keep selling them to editors and publishers. Exclusive terms can pay more, but they lock the archive to one buyer for the term, so price them for every other deal you give up. If you agree to exclusivity, limit it by field of use, time and image set.
Are old photos worth anything to AI companies?
Some are. Old photos with clear ownership, good scans and real captions can be worth more than new stock, because they show places, events and styles no one can shoot again. Loose snapshots with no rights trail are worth little to a buyer. Value comes from the collection as a whole: size, range of subjects, caption quality and proof that you own it.
Where can you sell old photos?
For single prints, collectors, auction houses and historical societies buy old photographs as objects. For licensing rights, stock agencies take contributors, and picture libraries license historical images to publishers. For AI training, buyers want whole collections with rights, captions and scans, licensed rather than sold. Whichever route you take, keep the copyright unless you mean to sell it, and check that you own it first.
Do AI labs buy images?
Yes, mostly through licenses rather than purchases, and mostly from large libraries. Agency filings show contributors being paid for inclusion in AI training sets, and AI search companies have signed multi-year image display deals. Labs prefer collections with clear ownership, releases where people appear, and good captions, because those cut their legal risk. A small archive can still license if its rights and metadata are clean.
Can a newspaper license the freelance photos in its archive for AI?
Only where its contracts give it that right. Under the copyright statute each contribution to a newspaper keeps its own copyright, and without a written transfer the paper is presumed to hold only the right to use it in that publication and its revisions. Freelancers at most major US papers still own their photos. Sort freelance images by contract and leave out those without a transfer or an AI license.
Do we need model releases to license news photos for AI training?
Copyright and consent are separate questions. An editorial archive can own the copyright in a crowd scene without any release from the people in it. Many buyers training image models still ask for releases, or for images where people are not the subject, and they avoid minors and private people in sensitive settings. Mark images with identifiable people and let the buyer choose what to take.
Can a photographer's estate license the archive for AI training?
Yes, if the estate holds the copyright. Check what the photographer licensed during their life: agency contracts, publisher deals and any exclusive grants still in force. Work made as an employee belongs to the employer, not the estate. A single photographer's archive with complete negatives, contact sheets and notes is often easier to clear than a newspaper's, because one person owned most of it.
Sources and further reading
- U.S. Copyright Office: Title 17, chapter 2 (17 U.S.C. 201 to 205), ownership and transfer
- U.S. Copyright Office: Title 17, chapter 1 (17 U.S.C. 101 definitions)
- Columbia Journalism Review: Exposure to Risk, freelance photographers and a new contract (July 8, 2026)
- Getty Images Holdings: Form 10-Q for the quarter ended September 30, 2026 (SEC)
- Getty Images Holdings: Form 10-K for 2025 (SEC)
- Getty Images: Getty Images and Perplexity strike multi-year image partnership (October 31, 2025)
- Getty Images: display partnership with OpenAI (June 21, 2026)
- Courts and Tribunals Judiciary: Getty Images v Stability AI, [2025] EWHC 2863 (Ch)
- Courts and Tribunals Judiciary: Getty Images v Stability AI, judgment of 4 November 2025 (PDF)
- Pinsent Masons Out-Law: Why Getty's appeal won't address all UK AI copyright questions (November 9, 2026)
- AP: Stability AI largely wins UK court battle against Getty Images (November 4, 2025)
- PetaPixel: Photobucket in negotiations with AI companies (April 8, 2024, reporting Reuters)
- VentureBeat: Apple's $25-50 million Shutterstock deal highlights competition for AI training data (April 8, 2024, reporting Reuters)
- U.S. Copyright Office: Copyright and Artificial Intelligence, Part 3, Generative AI Training (pre-publication version)
