Bria AI

The legal question that became a product

The biggest open question in generative AI is not technical but legal: what the models were trained on. Most of the leading image generators were trained on material collected from the web without explicit licensing, and many lawsuits are underway over this. Bria AI, founded in 2020, turned the answer to this question into the product itself. It licenses in advance all the content it trains on, and runs an attribution mechanism that tries to identify which sources influenced a generated image, so that royalties can be paid accordingly.

Getty Images as partner and investor

The main partner in this approach is Getty Images, the photo archive giant. According to Reuters, Getty has backed Bria since 2022, and it is also one of the platform's content sources. According to the company's official announcement, Bria has more than 30 content partners, including Envato, Alamy, Freepik and Depositphoto. The main customers are organizations and brands, exactly the ones exposed to lawsuits or reputational damage if it turns out that their marketing material was created on the basis of stolen work. That is why the product is sold as "safe for commercial use" and not as "the most beautiful."

Funding: from Series A to Series B

The funding rounds track the approach's progress: $24 million in a Series A in February 2024, with participation from the advertising agency Publicis Groupe. In March 2025 the company raised another $40 million in a Series B led by Red Dot Capital, bringing its total funding to about $65 million. "Since its inception, Bria's mission has been to bridge the accountability gap between organizations and Gen AI," said Dr. Yair Adato, Bria's founder and CEO, in the company's official announcement of the round.

Market direction or niche

The question that remains open is whether this is the market's main direction or a niche. Training only on licensed data limits the amount of data, and so may hurt quality compared with models trained on the whole web, and it is more expensive. As of 2026, the major lawsuits on the subject around the world have not yet been decided, and their outcome, not the quality of the models, is what will determine how much the approach the company chose is worth.

The technical challenge of attribution

Beyond the legal question, the attribution mechanism poses a hard technical challenge. It is very difficult to say with certainty which training images influenced a particular output, because a generative model does not store copies but statistical patterns. Every solution in this area is necessarily an approximation, and the quality of the approximation is part of the commercial claim.

What will happen to creators

The subject touches on a broader question the whole industry is dealing with: if creators are not paid for the material the models were trained on, what incentive will there be to keep producing that material in the first place? Licensing in advance is one of the proposed answers, alongside paying after the fact and relying on "fair use." All three approaches are competing with each other in courts and in legislation at the same time.