Is paying artists enough to convince them to embrace AI?

Illustrators have spent years objecting to generative AI startups that train models on artists’ work without permission. They describe that practice as theft. Boosters of generative AI have answered that scraping and training on existing art is necessary for the technology’s evolution. That clash has moved into contentious legal battles. The open question The Verge frames is whether paying artists is enough to get them to embrace AI rather than fight it.

The core product mechanic at issue is how generative models are trained: large volumes of existing images and styles are ingested so the system can produce new outputs that resemble those sources. Artists argue that using their work in that pipeline without consent treats their labor as free training data. Boosters treat that same pipeline as a requirement for model quality and progress. The dispute is not only about finished images; it is about whether the training step itself is legitimate when the source work was never licensed.

For engineers and builders, this is a data and consent problem, not only a content-moderation problem. Teams shipping generative tools inherit whatever rights story sits behind their training sets. If that story rests on “training is fair use or inevitable,” product, legal, and partnership risk all sit with the builder. If the industry shifts toward paid or licensed training data, build plans change: sourcing, model updates, and cost structure all depend on whether the data can be used cleanly.

The market side of the fight is between two incentives. Generative AI startups need scale of training data and speed of iteration. Working illustrators need control over how their portfolios are used and some form of compensation or permission when that work becomes model input. Paying artists is one proposed bridge: it treats training data as a cost of doing business instead of an unpriced input. Whether that is enough depends on whether artists accept payment as a substitute for control, credit, and the right to refuse use of their work.

What to watch next is whether compensation schemes actually reduce resistance or only paper over the permission gap. Watch for settlements, licenses, and product claims that specify opt-in or paid training data rather than blanket use of scraped art. For anyone building on generative image models, the practical check is simple: know what your model was trained on, whether that use was permitted or paid for, and what happens to your stack if courts or platforms force that answer into the open.