As technology companies delve into the commercialization of generative AI, the original creators upon whose work it is founded are advocating for fair compensation. However, there is currently no consensus on the appropriate remuneration for artists. A recent open letter from the Authors Guild, endorsed by over 8,500 writers, including prominent figures like Margaret Atwood, Dan Brown, and Jodi Picoult, implores generative AI companies to cease utilizing their works without proper consent or compensation. Concurrently, artists have initiated numerous lawsuits against generative AI providers such as Stability AI, MidJourney, and Microsoft, raising concerns about copyright infringement and misuse.
Some vendors have made commitments to establish “creators’ funds” and other mechanisms to remunerate the artists, authors, and musicians whose creations were instrumental in the development of their generative AI models. A few have even initiated these funds, touting them as steps toward more equitable and sustainable generative AI business models.
But what can creators realistically anticipate earning from these funds? This appears to be a straightforward question, but delving into the various compensation policies proposed by generative AI vendors reveals that it is exceptionally challenging to answer. We attempted to obtain clarity on this matter repeatedly, to no avail.
Generative AI models acquire the ability to generate images, music, text, and more by recognizing patterns in a vast number of examples, typically sourced from the publicly accessible internet. These examples, frequently encompassing copyrighted or licensed material, are often used without the creators’ knowledge. While some generative AI tool developers argue that they are justified in training on copyrighted works under the “fair use” doctrine, this issue is unlikely to be resolved soon. Moreover, public sentiment largely favors creators, who often earn significantly less compared to the substantial profits accrued by tech and AI companies.
Companies such as Adobe, Getty Images, Stability AI, and YouTube have introduced or pledged to introduce ways for creators to partake in their generative AI revenue. Nevertheless, these companies have not disclosed the exact amounts creators can expect to receive, making it a difficult decision for creators contemplating whether to allow a vendor to use their work for model training.
Adobe, for instance, trains its generative AI models, known as Firefly, using images from its Adobe Stock asset library. They state that they will provide an annual “bonus” that varies for each contributor, primarily based on the number of approved images used for training and the licenses generated by those images during a year-long period. Adobe, however, has not disclosed the specific value of each approved image or license. Contributors must also reach a $25 minimum threshold for withdrawals, with no guarantee of perpetual bonus payments.
Similarly, Getty Images plans to compensate contributors to its generative AI tool on an “annual recurring basis” but has not provided details on the compensation structure. Shutterstock, another generative AI tool provider, offers one-time payments through its Contributors Fund, with payments based on a creator’s contributions to Shutterstock’s content library. However, the exact proportions and additional compensation remain undisclosed.
In contrast, Stability AI’s revenue-sharing scheme for its Stable Audio model, created in collaboration with AudioSparx, is still being worked out, with no concrete earnings reports available yet. YouTube, in partnership with Universal Music Group, is in the early stages of developing monetization models that account for generative AI but has not provided specifics.
Regrettably, none of the generative AI vendors contacted were willing to disclose the average earnings creators can expect after contributing their works to model training. Some attributed the lack of data to the novelty of the technology and business model, while others argued that the range of earnings would be too diverse to provide a meaningful figure.
For creators, especially those reliant on contractual income, these explanations may seem unsatisfactory. Few vendors currently present a compelling case for artists to participate in generative AI model training, offering, at best, vague promises of future rewards that do not address immediate financial concerns.
