Shutterstock states that it compensates contributors whose work was used, Getty Images states that it compensates contributors on a recurring basis, and Credtent states an 85 % share, all as dated claims (checked November 9, 2026).
Artists are one of the groups within AI training data licensing, and a class action by artists against image-model developers is pending as of October 19, 2026. Licensing, reserving and doing nothing are all valid choices.
DataPacts Editorial Team checked every term and ruling on this page against its source, and the page on how we source, date and check every figure sets out the rules.
What Visual Artists Hold That AI Developers Train On
AI developers train image models on finished artworks together with the text that describes them, so an artist's portfolio and its captions are both of value. Wirestock states that it wants photos, video, illustrations, vector and 3D work, and Shutterstock's contributor fund covers images, video and metadata. The works in question are:
- finished illustrations and paintings
- concept art and sketches
- vector and 3D work
- captions, tags and metadata that describe the work
Some laws name this use directly. Singapore's Copyright Act 2021, section 243, gives "using images to train a program to recognise images" as an illustration of computational data analysis, and section 244 requires lawful access.
Artists are one of the groups among who sells data to AI companies, and a single artist usually holds far fewer works than a stock library.
Licensing Routes Open to an Individual Artist
An individual artist licenses through someone who aggregates many works, because no AI developer on record publishes a program for single portfolios. The table compares the routes and the terms each operator states, as of 2026-11-09, with intermediaries and collective licensing organizations shown as handling or licensing, not buying.
| Route type | Operator (role) | What it acquires or handles | Artist control as stated | Money terms published | Source |
|---|---|---|---|---|---|
| AI developer program | Bria AI (buyer) | licensed visual assets; trains only on licensed partner data | not published | "ongoing compensation" for works above relevance thresholds, tied to influence; no rate | bria.ai |
| Contributor app | Kled AI (buyer) | photos, video, music and other files uploaded by individuals | exclusive, perpetual, irrevocable, sub-licensable license; deletion on account closure not guaranteed | rewards in fiat or tokens at the company's discretion | kled.ai terms |
| Library licensing its own catalog | Getty Images | contributor images and video | under contributor agreements; no AI opt-out stated | recurring compensation, no rate | Getty 10-K FY2025 |
| Library licensing its own catalog | Shutterstock | contributor images, video and metadata | automatic for included content | compensation where contributor IP was used; no rate | submit.shutterstock.com |
| Licensing platform (intermediary) | Credtent | catalogs of rights holders; free opt-out notice | rights holder lists the catalog | states 85 % to rights holders under standard terms, quarterly | credtent.org |
| Licensing platform (intermediary) | Wirestock | photos, video, illustrations, vector, 3D | per project | states 50 % commission per license or a fixed payment per approved asset | wirestock.io |
| Collective license (UK) | Copyright Licensing Agency (collective licensing organization) | opted-in books, journals and magazines; no visual-art category stated | opt in via PLS or ALCS | not published | cla.co.uk |
| Collective license (US) | Copyright Clearance Center (collective licensing organization) | CCC repertoire for external AI training | not stated for artists | not published | copyright.com |
| Collective license (web) | RSL Collective (collective licensing organization) | web content of publishers and creators, declared in robots.txt | License line per site | royalties per use, rates not published | rslcollective.org |
Stock Libraries and Platforms That Pay Contributors
Stock libraries license their whole catalog and pay contributors whose works were included, while platforms let an artist choose per project or per use.
Shutterstock states that its Contributor Fund is automatic for contributors whose content is included and that first AI dataset earnings were issued in Q4 2022; the fund sits within the Shutterstock data licensing profile.
Getty Images states in its FY2025 10-K that it compensates contributors for training use and that partner models compensate creators on a recurring basis, with no rate.
Bria states that it trains only on licensed data from partners including Getty Images, Alamy and Envato, and that its compensation is tied to influence.
Wirestock offers per-project work, and Kled's license is perpetual and irrevocable, and Kled states it is exclusive to the company for training third-party AI models. Adobe's contributor terms for AI training could not be verified and are not stated here.
Collective Schemes: What Exists for Visual Art
No collecting society for visual art has a published AI training license on record, so the collective routes that exist reach artists only indirectly. The CLA's Generative AI Solution covers opted-in books, journals and magazines, with publishers opting in via PLS and authors via ALCS, and the CCC and RSL Collective publish no payout.
The US Copyright Office's Part 3 report (pre-publication version of 2025-05-09, final not published as of 2026-11-09) recommends letting the licensing market develop and considering extended collective licensing where market failures are shown.
In the UK consultation report of March 2026, 81 % of 11,520 responses preferred option 1, which strengthens copyright and requires licensing. Programs for other repertoires are explained under collective licensing for AI training.
What Artists Are Paid for AI Training: Dated Claims Only
No route publishes a rate per image or per artwork, and the figures that exist are each company's own claim or an aggregate in a filing. The table lists the stated terms, as of 2026-11-09, with Getty's filing values as of 2025-12-31.
| Company | What is stated | Basis | Stating party and date | Source |
|---|---|---|---|---|
| Bria AI | compensation tied to influence, not catalog size; no rate | program page | Bria, checked 2026-11-09 | bria.ai |
| Credtent | 85 % to creatives, rights holders and data providers, quarterly | standard terms | Credtent, checked 2026-11-09 | credtent.org |
| Getty Images | over USD 220 million royalties to contributors in 2025, all products, not AI alone | annual report | Getty 10-K FY2025 | sec.gov |
| Shutterstock | contributors compensated where their IP was used; first AI dataset earnings Q4 2022; no rate | contributor FAQ | Shutterstock, checked 2026-11-09 | submit.shutterstock.com |
| Wirestock | 50 % commission each time an asset is licensed; fixed payments per approved asset | creator blog | Wirestock, checked 2026-11-09 | wirestock.io |
Three cautions apply to every row:
- Getty's total covers all products and more than 600,000 contributors, over 83,000 of them exclusive, so it is not an AI-only figure; Getty's "Other revenue", which includes data access and licensing, was 5.6 % of 2025 revenue.
- A share is a share of the platform's receipts, not of a buyer's price.
- No figure is a market price.
Getty filed for Chapter 11 on 2026-10-21 (Bankr. D. Del. 26-11643 (JKS)), and its first-day motions state that contributors are paid in the ordinary course, as of 2026-11-09. Licensing income from works an artist created is generally ordinary income under 26 U.S.C. 1221(a)(3), which is covered under tax on selling your own data as an individual or creator.
Reserving a Portfolio Against Training: Signals Courts and Crawlers Read
In the EU an artist's reservation against commercial text and data mining counts only when it is machine-readable, which a sentence on a website may not be. Directive (EU) 2019/790, Article 4(3), requires a machine-readable reservation for online content, and Article 7(1) protects the research exception in Article 3 against contract override, not Article 4.
In Kneschke v. LAION, the OLG Hamburg (5 U 104/24, 2025-12-10) dismissed the appeal because the claimant had not shown that the agency's natural-language reservation was machine-readable in the second half of 2021, and held the use additionally covered by the research exception, and the Federal Court of Justice (I ZR 281/25) is scheduled to pronounce on 2026-12-17. The signals on record are:
- a robots.txt rule: OpenAI states that disallowing GPTBot signals that content should not be used in training (checked November 9, 2026), and Measure 1.3 of the GPAI Code of Practice (2025-07-10) has voluntary signatories follow robots.txt under RFC 9309
- the cawg.training-mining assertion, version 1.1, for image files; C2PA removed its own assertion in version 2.0
- a notice sent through a service: Credtent states that its opt-out is free and followed up at 30, 60 and 90 days
Each signal is compared under machine-readable TDM reservations. A reservation does not reach copies already made.
Lawsuits Brought by Visual Artists Over Training Data
Andersen v. Stability AI is the class action brought by visual artists over image-model training, and as of October 19, 2026 it is in discovery.
The complaint, filed 2023-01-13 in N.D. Cal. (3:23-cv-00201-WHO), names Sarah Andersen, Kelly McKernan, Karla Ortiz and other artists as a putative class against Stability AI, Midjourney, DeviantArt and Runway AI, and alleges copyright infringement through LAION-based training; the case page for Andersen v. Stability AI holds the detail. The dated rulings are:
- 2024-08-12 (Dkt. 223): the DMCA claims were dismissed with prejudice and unjust enrichment with leave to amend
- 2024-08-12 (Dkt. 223): the Copyright Act claims survive, and so does the Lanham Act claim against Midjourney
- 2026-10-19 (Dkt. 778): a stipulation and order resolved the sanctions motion against DeviantArt
As of October 19, 2026 the docket shows discovery, a plaintiffs' motion of 2026-09-23 to modify the schedule and no trial date; no later entry was found in the docket archive on November 16, 2026. The claims are allegations, and nothing has been decided on the merits.
In the UK, Getty Images sued Stability AI over image-model training. The High Court ([2025] EWHC 2863 (Ch), 2025-11-04) held that the model is not an infringing copy and found trade mark infringement in part, and the Court of Appeal (CA-2026-000007) began hearing the appeal on 2026-11-11 and reserved judgment on 2026-11-12 (case tracker, checked November 16, 2026). The appeal is followed under Getty Images v. Stability AI in the UK courts.
Rights an Artist Must Hold Before Licensing
An artist can license only works whose copyright they still own and have not licensed exclusively to someone else. Three groups of works need checking first:
- commissioned work and work made for hire
- work made as an employee
- work already licensed exclusively to a client or library
Under 17 U.S.C. 201(b) and 101, a commissioned work is made for hire only in nine listed categories and with a signed written agreement, so the client may own it.
Under section 11(2) of the UK Copyright, Designs and Patents Act 1988, the employer is first owner of an artistic work made in the course of employment, subject to agreement.
Exclusivity matters too: Getty states that over 83,000 of its contributors are exclusive. The general method is the rights audit before licensing data, which covers chains of title.
Licensing, Reserving or Doing Nothing: Benefits, Risks, Downsides and Alternatives for Artists
Licensing earns money only through an aggregator's terms, and reserving or doing nothing keeps control but pays nothing. The lists below set the options side by side.
Benefits
- Compensation where work was used, as Shutterstock and Getty state, and ongoing compensation tied to influence, as Bria states
- A stated 85 % share of platform receipts, as Credtent states, paid quarterly
- No cost to send an opt-out notice through Credtent
Risks
- An exclusive, perpetual, irrevocable license such as Kled's cannot be undone
- An intermediary can enter insolvency, as Getty did on 2026-10-21
- The outcome of pending litigation is unknown as of 2026-10-19
Downsides
- No AI-only figure exists, so earnings cannot be forecast
- A reservation that is not machine-readable may not count in the EU
- The UK exception in section 29A covers only non-commercial research, so UK artists have no commercial exception to rely on
Alternatives
- A platform with per-use opt-in, such as Wirestock's projects or Credtent
- A machine-readable reservation under Article 4(3)
- Following the pending class action
- Not licensing at all, or doing nothing
The general decision is set out under should you sell or license your data.
How Visual Artists Differ From Other Creative Sellers
Visual artists share routes with photographers, video creators and stock contributors, but they hold fewer works each and rarely sell whole archives. Programs for creators selling video, photo and writing are compared separately. Related pages:
- selling photos and images for AI training
- AI training settings on platforms and their limits
- musicians and labels licensing music
- stock photo and footage libraries licensing to AI
- consent and compensation for creators
Case, Law and Program Pages Named on This Page
The rulings, laws and program terms on this page are each dated to the source that states them. This page is information, not legal advice.
- AI copyright lawsuits over training data
- US Copyright Office report on generative AI training
- EU text and data mining exception
- Kneschke v. LAION
- LAION-5B
- Wirestock and Bria AI program profiles
- training data attribution and per-use payment
Frequently Asked Questions
Do AI companies pay artists whose work was used for training?
Only through the terms of a library or platform that licensed the work; no AI developer publishes a per-artwork rate. Shutterstock and Getty state compensation without a rate, and Bria ties compensation to influence (checked November 9, 2026).
How can an artist keep their work out of AI training?
An artist can publish a machine-readable reservation and block training crawlers, but neither reaches copies already made. Article 4(3) of Directive (EU) 2019/790 requires machine-readable form, and OpenAI states that disallowing GPTBot signals no training use.
Has Andersen v. Stability AI been decided?
No; as of October 19, 2026 the case is in discovery, and the copyright claims survived a 2024 motion to dismiss. The order of 2024-08-12 (Dkt. 223) dismissed the DMCA claims with prejudice.
Can an illustrator license work through a stock library?
Yes, if the library accepts illustrations and pays contributors for AI licensing, as Shutterstock's fund and Wirestock's projects state. Wirestock states that it wants illustrations, vector and 3D work (checked November 9, 2026).