A stock agency priced indemnity into its own generator

The process wall

On 25 September 2023, Getty Images said it would sell access to a generator built on its own catalogue rather than only licence photographs made by people. The announcement frames this as a way to give customers image generation without the legal exposure attached to tools trained on scraped web images. That framing rests on a sourcing decision a photo library makes before any prompt is typed, not a stylistic one made afterwards.

What the documents show

The announcement states the tool is trained solely from Getty Images' vast creative library, built on NVIDIA's Edify architecture, and that generated visuals ship with Getty's standard royalty-free licence, which the document says carries representations, warranties and uncapped indemnification for commercial use. It also states that images made with the tool will not return to the Getty or iStock libraries for others to licence, and that contributors will be compensated for any inclusion in the training set. The current product page, reflecting the service as retrieved on 16 September 2026, ties that compensation to usage: each generation draws a credit from a purchased agreement, funding ongoing payments to the people whose work trained the model, and states legal protection starting at $50,000 USD per generated image for the buyer.

Authorship and limitation

'Trained solely' and 'commercially safe' are Getty's own claims about its own catalogue, not an independent audit, and they hold only as far as Getty's existing contributor terms already permitted this use — the announcement describes payment for content already under contract, not a new opt-in vote. The mechanism worth separating is licensing from authorship: Getty can warrant that it holds rights to its training inputs, but that warranty is a contractual promise to indemnify a buyer, not a claim that a generated image is an authored work belonging to whoever typed the prompt. Indemnity also has a stated boundary — the current page's figure is a starting at price, meaning coverage is tiered, and neither document says what a claim beyond that tier would leave a customer holding.

What survives

A maker working from a Getty generation keeps a licence with named warranties rather than a bare file, a different asset from a hand-made image with a clear authorship chain. Editorially, anyone relying on this for client work should confirm which licence tier and indemnity figure attaches to a given export, since 'commercially safe' describes a training claim, not a property of the pixels. Contributors should check their own contributor terms, not the product page, for the exact calculation basis.

• Read the specific licence text attached to a Getty-generated export before reusing it commercially.
• Ask Getty's contributor support how 'inclusion' in the training set is defined and calculated.
• Compare the advertised indemnity figure against the plan actually purchased.

The two documents describe a vertically integrated bet: a library holder selling both the training data's provenance and the legal cover that provenance is meant to justify, with contributor payment folded in as a condition of using its own catalogue rather than a separately negotiated right.