As tech companies begin to monetize generative ai, the creators whose work is trained are asking for their fair share. But so far no one has agreed on whether artists should be paid and how much they should be paid.
A recent open letter from the Authors Guild signed by more than 8,500 writers, including Margaret Atwood, Dan Brown, and Jodi Picoult, calls on generative ai companies to stop using their works without proper authorization or compensation. Meanwhile, artists have filed numerous lawsuits against generative ai vendors such as Stability ai, MidJourney, and Microsoft over copyright and misuse.
Some vendors have committed to establishing “creator funds” and other means to pay artists, authors and musicians whose works they have used to develop their generative ai models. Some have even taken the step of in fact launch said funds, which they have heralded as a move toward more equitable and sustainable generative ai business models.
So how much can creators realistically expect to earn from these funds?
It seems like a simple question. But when you dig deeper into the various compensation policies proposed by generative ai vendors, it becomes exceptionally difficult to answer. Trust us, we try. Repeatedly.
Vague terms
Generative ai models “learn” to create images, music, text, and more by capturing patterns in a huge number of examples, usually obtained from the publicly accessible web. The examples (usually photographs, artwork, audio, and text) are often copyrighted or released under a usage license that the providers are unaware of, and the creators are often not even informed that their works are being used in any way. this way.
While some companies developing generative ai tools argue that training on copyrighted works under the “fair use” doctrine is justified, at least in the US, it is an issue that ai-copyright-violation-training-data” target=”_blank” rel=”noopener”>unlikely which will be resolved soon. And legal issues aside, public opinion has largely rallied behind creators, most of whom earn a pittance compared to the billions being raked in by tech and ai companies.
That’s why vendors like Adobe, Getty Images, Stability ai, and YouTube have introduced, or promised to introduce, ways for creators to share their generative ai profits. The problem is that companies haven’t made it clear exactly how much creators can expect to earn. And for creators considering allowing a vendor to train a model on their works, the decision isn’t easy.
Adobe, which trains its family of generative ai models, called Firefly, on images from its Adobe Stock asset library, says it will pay an annual “bonus” that is “different for each contributor.” The first was disbursed at the beginning of September.
The Adobe bonus is based primarily on the total number of approved images, vectors, or illustrations submitted to Adobe Stock Standard or Premium that were used for Firefly training and the “number of licenses” your images generated over a one-year period. year, a spokesperson told me. via email. Future bonuses will be calculated from new images and approved downloads, meaning creators can’t rely on metrics in a previous bonus period to predict their next payout.
Image credits: Adobe
How much is each individually approved image and license worth? Unclear. Adobe refused to inform us.
The only thing we know for sure is that taxpayers must meet a minimum threshold of $25 before they can make a withdrawal (with the exception of taxpayers who received the first bonus payment, who can withdraw at $1 between September 13 and December 12th). Adobe says it can take 8 to 10 business days or more to complete a withdrawal. And, alarmingly for taxpayers, the company does not guarantee that it will pay bonuses in perpetuity.
But wait, it gets more complicated and opaque.
The Firefly bonus is currently weighted by the number of licenses issued for an image, the Adobe spokesperson said, which the company considers an indicator of demand and the “usefulness” of an image. But Adobe would not say to what extent it is weighted and whether the weighting will change in the future.
Getty Images also plans to pay contributors to its recently announced generative ai tool on an “annual and recurring basis,” according to a spokesperson. Content creators will get a “prorated” (i.e. proportional) share for each asset they contributed to the model training data set, as well as a share based on “traditional licensing revenue.”
We requested clarification on the licensing portion and more information on the prorated payment agreement. However, like Adobe, Getty Images was not forthcoming about the details.
“There will be a set formula based on a number of different factors and consequently each taxpayer will receive different payments in relation to the tool,” the spokesperson said.
Image credits: fake images
Getty Images competitor Shutterstock, which also offers a suite of generative ai tools and sells its metadata and stock images to partners like OpenAI, distributes one-time payments through its Contributors Fund. Semi-annual payments are proportional to a creator’s contributions to Shutterstock’s content library, and creators receive additional compensation if new content produced by Shutterstock’s ai generators includes their work.
“Contributors will receive a portion of the total contract value paid by customers who license datasets,” Shutterstock writes on their website. “Contributors whose content was used to train (models) will be compensated for the role their intellectual property played in the development of the original models, as well as through royalty payments tied to future generative licensing activities.”
What is the exact proportion? And what would that “additional compensation” look like? It’s a mystery.
The best estimate we have is that of archival photographer Robert Kneschke, who was responsible for ai-contributor-fund/#:~:text=Shutterstock%20May%20Have%20Paid%20Out%20Over%20%244%20Million%20From%20its%20AI%20Contributor%20Fund,-Jul%2012%2C%202023&text=Shutterstock’s%20Contributor%20Fund%20aims%20to,the%20company%20is%20currently%20promoting.”>survey 58 photographers plus how much they were paid from the Shutterstock Contributor Fund and take their portfolio size into account when calculating averages.
Kneschke’s survey found that the average income from the Taxpayers Fund was $0.0078 per image, while the median was $0.0069 per image. Assuming those figures are accurate, a photographer with around 2,000 images would earn approximately $15, not exactly a huge amount.
No dollar amount
Incredibly, those are the most concrete generative ai compensation schemes we could find. The others are more… theoretical.
When Stability ai announced Stable Audio, a model that generates music and sound effects from a text description, the ai startup said that, through its partnership with audio library AudioSparx, it would allow musicians to share in the profits. generated by Stable Audio. All they would have to do is join AudioSparx and choose to participate in training the initial model or decide to help train future versions of Stable Audio.
A few weeks later, the details of that revenue-sharing plan are still being discussed, according to Lee Johnson, executive vice president of AudioSparx.
“We haven’t received any earnings reports from Stability ai yet, and we’re still in the early days in terms of understanding the revenue that will be generated,” Lee told TechCrunch. “As such, it remains to be seen what kind of gains the average taxpayer can expect to make.”
Image credits: ai Stability
Lee went on to say that contributors can expect to receive a portion of the profits generated by Stable Audio on a “residual and recurring” basis as long as they choose to participate in model training.
“Once we receive the first earnings report from Stability ai and are able to fully understand the various metrics and details of the information they will provide, we will have the necessary information on hand to fully determine how to allocate earnings to each. of the participating artists,” Lee said. “There is an ongoing discussion between AudioSparx and Stability ai regarding some of the issues surrounding earnings metrics and reporting, so this is all still in development.”
Elsewhere on the generative ai music front, YouTube, which in August unveiled a generative ai partnership with Universal Music Group, said it plans to develop a structure that ensures music rights holders are paid for their contributions. of training data. But when contacted for content, YouTube said it is in the “early days” of creating monetization models that take generative ai into account.
“A big part of that will be achieved by collaborating with our partners across the music business,” a YouTube spokesperson said.
Bad luck, creators.
Tellingly, none of the generative ai vendors we spoke to would give the amount of dollars the average creator can expect to see after shelling out their creations for model training.
Some providers blamed the lack of data on the newness of the technology and business model. Others said the range would vary too much to give a useful figure.
But for creators, particularly those who rely on contract revenue to make ends meet, those are arguments that probably ring hollow.
Some startups are trying to be more transparent (and creator-focused) from the start. Braia, which trains its art-generating ai strictly on licensed images, has a revenue-sharing model that rewards data owners based on the impact of their contributions, allowing artists to set prices per training run. of ai.
However, as far as we know, as things stand now, few vendors are making especially compelling arguments that it will be worth it for artists to opt for generative ai model training. At best, they are offering confusing promises of future riches, and those confusing promises don’t pay the rent.
