Where Photographers And Illustrators Stand On AI Training Data
The AI training data argument is usually staged as artists versus technology companies, which is not the version that reaches a brand. The version that reaches a brand is narrower and more immediate: what can be published without acquiring a problem.
Advertising has always been careful here, because it is commercial use and commercial use is where rights are enforced. The industry built stock licensing, model releases and clearance processes precisely to make that risk manageable.
Why AI Training Data Sits Outside That System
Every one of those mechanisms works by tracing an asset to a source and a permission. Generated output has no such chain. It resembles its training corpus in ways that are real but not itemised, so there is nothing to point at and nothing to clear.
Vendors have responded with indemnities of varying strength. An indemnity is a promise to pay if it goes wrong, which is not the same as it not going wrong, and the reputational half is not indemnifiable.
The Working Position That Has Emerged
Larger advertisers have converged on something practical: use models trained on licensed or owned material for anything published, and treat the general-purpose tools as fine for internal work, boards and exploration.
The split is not ethical, it is procedural. It maps to which assets a legal team will be asked to defend.
What This Means For The People Being Argued About
Commissioned photography and illustration are not being replaced evenly. What generative tools take first is the low-margin, high-volume work that funded the rest, which is how a profession thins from the bottom rather than the top.
Brands that want distinctive imagery will keep commissioning, because generic is the one thing a general-purpose model reliably produces. Read with the sameness problem, the two are the same observation from opposite ends.
Advertisement
