As technology companies seek to monetize generative AI, creators whose works fuel these innovations are advocating for fair compensation. Yet, consensus on whether—or how much—artists should be paid remains elusive.
A recent open letter from The Authors Guild, signed by over 8,500 writers including notable figures like Margaret Atwood, Dan Brown, and Jodi Picoult, calls on generative AI companies to stop using their works without proper authorization and remuneration. In parallel, artists have filed numerous lawsuits against generative AI providers such as Stability AI, Midjourney, and Microsoft, claiming copyright infringement and misuse of their intellectual property.
In response, some companies have committed to launching “creators’ funds” and other initiatives aimed at financially supporting the artists, authors, and musicians whose works contribute to their generative AI models. A few have even implemented these funds, presenting them as steps toward more equitable and sustainable business practices in generative AI.
But how much can creators realistically expect to earn from these creators’ funds?
While it seems like a straightforward question, the complexity of various compensation policies proposed by generative AI vendors makes it anything but simple. We’ve looked into this matter extensively.
Vague Terms and Compensation Structures
Generative AI models develop the ability to create images, music, text, and more by identifying patterns in vast datasets often sourced from publicly accessible websites. These datasets typically include copyrighted material that creators have not consented to be included— a fact that many vendors overlook.
Some companies contend that training on copyrighted works falls under the “fair use” doctrine, particularly in the U.S. However, this issue remains hotly debated, and with public sentiment largely supporting creators—many of whom earn a small fraction compared to the profits made by tech giants—the path to resolution appears long and fraught.
Companies like Adobe, Getty Images, Stability AI, and YouTube are introducing—or promising—mechanisms for creators to participate in the profits linked to their generative AI technologies. However, these companies have yet to clarify the specific earnings creators can anticipate, complicating the decision-making process for those considering allowing their work to be used for AI model training.
For instance, Adobe, which trains its suite of generative AI models, Firefly, using images from its stock library, has initiated annual “bonus” payouts that vary for each contributor. The first payment was distributed in early September.
Adobe’s bonus system is primarily based on the number of approved images, vectors, or illustrations contributed to Adobe Stock and their respective licensing revenues over a one-year timeframe. Future bonuses will be assessed on new submissions, meaning past metrics won’t guarantee future earnings.
Unclear Value Assessments
What’s the monetary value of each approved image and license? Unfortunately, that remains unclear, as Adobe has refrained from providing specific figures.
The only certainty is that contributors need to meet a $25 minimum threshold before they can withdraw funds (participants from the initial bonus distribution have an exception allowing them to withdraw at a $1 threshold between September 13 and December 12). Adobe states that withdrawals can take eight to ten business days or more, and notably, there’s no assurance that bonuses will continue indefinitely.
The situation complicates further; a spokesperson indicated that the Firefly bonus structure currently weights compensation based on the number of image licenses issued, which Adobe interprets as a measure of demand and usefulness. However, the specifics of this weighting and potential changes remain undisclosed.
Similarly, Getty Images has announced it will compensate contributors annually based on a proportional share of assets contributed to its generative AI training dataset, alongside revenue from traditional licensing. While the company acknowledges a formula will be established to determine payments, specifics remain vague.
Transparency Challenges
Shutterstock operates a Contributors Fund that offers biannual one-time payments to reward content creators based on their contributions to the platform. They also provide additional compensation if any new content generated by their AI uses a contributor's materials. However, precise percentages and details about the compensation model are elusive.
In a survey conducted by stock photographer Robert Kneschke, it was estimated that the average compensation from the Shutterstock Contributors Fund amounted to about $0.0078 per image, leading to approximately $15 earnings for a contributor with around 2,000 images. That figure raises concerns regarding the adequacy of such payments.
For Stability AI’s newly introduced music-generating model, Stable Audio, the revenue-sharing framework with stock audio library AudioSparx is still under development. Contributors are expected to receive a share of profits, but specifics such as potential earnings remain to be seen as discussions continue on the model's metrics and reporting.
Conversely, YouTube, in partnership with Universal Music Group, claims it is in the nascent stages of creating monetization models that ensure music rights holders are paid for their contributions. However, clarity on the payment structure remains lacking.
Conclusion: The Need for Clear Compensation Models
None of the generative AI companies consulted provided a definitive estimate of earnings for creators sharing their work for model training. Some attribute the ambiguity to the newness of the technology and business structure, while others point out that the potential earnings could vary significantly.
For creators, especially those relying on contractual income, these justifications may feel unsatisfactory.
Emerging startups like Bria are attempting to establish transparency by training AI solely on licensed images, offering a revenue-sharing model that quantifies rewards based on contributions. Yet, under current conditions, few vendors convincingly demonstrate that participation in generative AI model training will yield worthwhile compensation for artists. The future remains uncertain, as vague promises cannot cover the reality of rent and living expenses.
Ranking
