Who Still Makes the Originals That AI Learns From?
As generative tools become part of the creative process, I keep thinking about the work that happens before an image ever reaches the screen.
A photograph can look simple when you only see the finished frame. Somebody chose the location, found the light, worked with the person in front of the camera and decided when to press the shutter. Then came the quieter choices: Which photograph holds up? What gets cropped? Which details should stay?
I think about that when I see new AI tools producing convincing images almost instantly. There is a whole collection of photographs, illustrations, recordings and other work behind the systems we use. How do we recognize the people who made those originals, especially as the finished output becomes easier to produce?

The work behind an image
Imagine a photographer planning a portrait. The location might be available for only an hour. The subject has their own expectations. The photographer has to build trust, read the light and adjust when something unexpected happens. The image that eventually makes it into a portfolio carries some of that work, but the file doesn't explain the entire process.
That experience shapes what the photographer knows how to see. It also creates something another artist can study, reference or license. The same is true of a designer who spends days making a garment, or an illustrator who learns a particular way to draw an expression.
Generative systems introduce another possible use for original work: training. That raises questions about permission and payment, although the facts differ considerably depending on the model and the material involved. We should resist treating every company or every artist's experience as the same.
One company offers a revealing example
Adobe says its first commercially released Firefly model was trained on Adobe Stock images, openly licensed material and public-domain work where copyright had expired. The company also says that eligible Adobe Stock contributors received a Firefly bonus on Sept. 17, 2026. According to its contributor guidance, the amount varies and is paid at Adobe's discretion. The 2026 calculation considers eligible content used for training during a defined period and the number of licenses those assets generated. Adobe says Stock contributors currently do not have an option to exclude their Stock content from AI training. Adobe's Firefly contributor FAQ
I find that arrangement worth examining because it gives us something specific to discuss. There is a licensing relationship and a payment mechanism. But the existence of a bonus tells us little about whether the payment is meaningful to any particular photographer. We would need actual earnings and contract terms, and different contributors may reach different conclusions.
Getty Images provides another model. In its public financial filings, the company says that many contributors retain ownership of their material while granting licensing rights, and that contributors may be compensated when their content is included in AI training sets. Its reported terms depend on the type of license and contributor arrangement. These are company disclosures, not evidence that every participant earns a sustainable income. Getty Images' SEC filing
What about work that was never offered for training?
That question is still being argued. The U.S. Copyright Office released a prepublication report on generative AI training on May 9, 2025. It examines the legal and policy questions surrounding the use of copyrighted material. That report does not resolve every specific use, and a broad claim that all training is either permissible or infringing would go beyond what we can establish here. U.S. Copyright Office AI reports
Creative Commons has also pointed out that copyright permissions alone cannot settle the questions of consent, privacy, representation and economic effects. An artist might be willing to let people learn from a work, yet disagree with a particular commercial reuse. An open license has actual terms; it is not an invitation to ignore every interest connected to the subject. Creative Commons on generative AI

Making something and choosing what survives
There is another part of creative work I don't want us to overlook: selection. Two people can work from the same subject and finish with photographs that feel completely different. One might be interested in movement, another in stillness. One gets close, another gives the surroundings room to tell the story.
Those choices develop from looking at other people's work, from trying things and from discovering what you care about. As technology expands the number of images we can produce, I wonder whether we will spend more time studying the origins of those choices or less.
I also wonder what happens to a beginner who once learned by assisting a photographer, setting lights or sorting through imperfect frames. Those tasks can be tedious. They also provide opportunities to notice things that don't appear in a prompt. We need better evidence before claiming that AI will eliminate those opportunities or create enough new ones to replace them.
The U.S. Copyright Office's separate January 2025 report on AI-generated outputs says protection depends on sufficient human expressive contribution. It makes room for AI-assisted work, while treating prompts alone differently from human-authored expression. That is a legal analysis of copyrightability, not a ranking of artistic value. Copyright Office report announcement
A question to carry into the next project
For me, one useful habit would be to keep a record of where the references came from. Who made the photograph? What was it made for? Was it licensed, purchased, shared with permission or simply found while browsing? And what part of that work actually influenced the result?
This doesn't answer the larger debate about training. It does make the people and decisions behind our references harder to forget.
The more tools we have for producing images, the more interested I become in the people who continue to go out and make the originals. What do they get to create next? How can they keep making it? I think those questions deserve as much attention as the next product demonstration.
Photographs: Jakub Żerdzicki and Vitaly Gariev, via Unsplash. Images are illustrative and are not presented as evidence that the photographers' work was used in AI training.




Comments