Copyright & creators

The image model that draws in a famous artist's style learned it somewhere. That 'somewhere' is millions of real creators' work — and they're asking whether anyone should have asked first.

Trained on human work

Generative models learn by studying enormous collections of existing images, text, music, and code — most of it made by people, much of it scraped from the open web. The results can echo styles and phrasings from that training set. Creators argue their work was used to build a competing product without consent, credit, or pay. AI companies argue that learning patterns from public work is fair. Both sides are in court.

Who owns the output?

Two questions get tangled. One is the training-data fight above. The other is who owns what the AI makes: in the US, copyright rewards human authorship, so a raw prompt-to-image output often isn't automatically yours. Rules differ by country and are still moving. For anything commercial, check the tool's licence and add real human creativity — don't assume a raw AI output is yours to sell. (General explanation, not legal advice.)

The takeaway

• —Generative AI is trained on huge amounts of human-made work.
• —Whether that training needs consent or pay is being fought in court.
• —Owning the output usually needs real human authorship — and rules vary.