Data licensing: when your stock footage trains AI

How stock footage AI training data licensing works

A dataset is a set of files and their metadata, put together on a theme for a company that trains a model. Shutterstock describes its datasets as visual content organized by topic, such as food and beverage or transportation and autonomous vehicles. Pond5 says its datasets serve companies building computer vision and large language models.

The two agencies describe the license in almost the same words:

• the buyer may use the content and metadata only to train machine learning and computer vision models;
• commercial or public use, such as marketing or advertising, is prohibited;
• the buyer must have security measures against unauthorized distribution.

Your metadata travels with the clip. Both list descriptions and keywords, Shutterstock adds categories, and both mention some geolocation and broad demographic labels about models, such as age, gender and race or ethnicity. Both say full model releases are never shared and that the identity of models and contributors is not disclosed. Shutterstock adds labels from its own machine learning models.

What each agency publishes

On the other side of the coin, Shutterstock, Pond5 and Getty Images refuse AI-generated content from contributors, and Adobe Stock accepts it with a label. That policy is covered on the AI video policies page.

Shutterstock: data licensing and the Contributor Fund

Shutterstock launched datasets in July 2021, and its help article says including existing library content is covered by section 1a of its Contributor Terms of Service. Datasets can hold photos, illustrations, vectors, video, 3D models and music. Some editorial content can be included; premier editorial content is excluded.

The Data Catalog: clips accepted for data licensing only

Since June 2023, Shutterstock can accept content specifically for datasets, and shows it in a Data Catalog in your account. Its quality requirements article (17 July 2025) applies this to review: a video that meets every legal, compliance and metadata rule but fails the minimum quality bar “will only be accepted for data licensing.” If you have opted out, such a clip is marked in the Review tab and Shutterstock keeps that result in case you opt in later. The quality bar itself is explained in video quality rejections.

Turning it off

1. Open Account settings and choose Edit at the top of the page.
2. Verify your account when asked.
3. In Licensing options, set data licensing to “no” for images, video or both.

Changes can take up to 72 hours. The 30-day wait before reinstating applies when you remove content from creative marketplace licensing, not data licensing.

Pond5 Dataset Earnings

• Contributors receive a 20% revenue share for each licensed item.
• Pond5 warns that license prices are lower than regular marketplace purchases, because the rights are restricted to a special purpose.
• All media types are eligible; what goes into a dataset depends on the client.
• Some Pond5 datasets are licensed with Shutterstock and include content from both.
• Dataset Earnings show as their own line in the Financials section, and you get an email for $25 or more in a period.
• Payments come when royalties are available, not on a fixed monthly or quarterly schedule.
• Since April 2023, an option under Preferences lets you opt out of future datasets. Earnings from deals closed before you opt out can still appear.

The rest of Pond5’s payout rules, from the 30% video share to the $25 threshold, are on the Pond5 requirements page.

Adobe Stock and Firefly

The Adobe pages cited here describe no dataset sales to outside buyers; they describe the training of Adobe’s own Firefly models. Its Firefly FAQ, updated on 16 September 2026, says:

• the first commercially released Firefly model was trained on Adobe Stock images, openly licensed content and public-domain content;
• Adobe Stock video is now used to train Firefly models;
• Adobe’s existing terms with contributors permit it to create and train Firefly models;
• Adobe does not offer an opt-out for Adobe Stock content.

How Adobe compares with Pond5 on this point, and on everything else, is in Adobe Stock vs Pond5.

Wirestock dataset deals

Wirestock’s Terms of Use, effective January 20, 2026, enroll every item submitted after that date in its Dataset Deals Program. The terms say there is no option to opt out: if you do not want a file included, do not submit it. They also stop sending new content to stock marketplaces from that date.

The usual payment is your fraction of the participating content multiplied by 50% of what the dataset partner pays; Wirestock keeps the other 50%. The terms give an example: 100 of your items out of 10,000 in a $10,000 deal pays you $50. For bundled deals, Wirestock sets the amount. Licenses already granted, to dataset partners included, survive if you close your account. The full picture is in Wirestock vs BlackBox.

Opting in or out: what changes

Three points hold across agencies:

• Opting out looks forward. Shutterstock and Pond5 both speak of future datasets. A clip already licensed in a dataset stays licensed.
• No agency publishes an amount per clip. Shutterstock gives an average rate, Pond5 a share of lower prices, Adobe a discretionary bonus.
• Your keywords go with the clip. Accurate metadata is part of what a dataset buyer pays for, which is one more reason to describe only what the frame shows.

How data payments fit next to regular sales is covered in how much stock footage earns.