The standard argument goes like this: AI companies should pay artists when their work trains generative models. Kapwing actually tried it. The results are a cold splash of reality for anyone who thinks royalties alone can solve the AI-art standoff.
In May 2024, Kapwing launched Tess, a marketplace where artists could license their style for AI image generation. The company fine-tuned Stable Diffusion models (the open-source image generation system) on individual artists' portfolios and listed them in a public marketplace. Artists received 50% of revenue when subscribers used their specific model. Founding artists got advance payments of $300 to $4,000.
Twenty months later, Kapwing shut Tess down. The final numbers are stark.
$18,000 Out, $12,172 Back
Kapwing paid $18,000 in advances across 25 founding artists. Total revenue from customers: $12,172.33 from 142 paying users. Infrastructure ran about $100 per month. Not a single artist earned enough from actual usage to exceed their advance payment. The program lost roughly $7,000 in direct costs, not counting engineering time.
The demand simply wasn't there. 142 customers over eight operational months works out to fewer than 18 new customers per month. For context, Kapwing's main product serves millions of users. Tess barely registered.
The Recruitment Problem Was Even Worse Than the Revenue Problem
Kapwing sent 325 cold emails to recruit artists. Half responded. Only 6.5% agreed to participate and have models trained on their work. Nearly a quarter gave hard rejections. Eleven illustration agencies were contacted and every single one said no.
The reasons artists refused fell into four buckets: ideological opposition to AI art in general, concern about diluting their brand, philosophical objections to machine-generated creativity, and fear of reputation damage from being associated with generative AI. A meaningful fraction of creators will refuse on principle regardless of compensation structure.
Even among artists who joined, engagement was low. Only about one in four ever used the artist dashboard to monitor how their models were being used.
The Legal Wall
Perhaps the most revealing detail: at least one major U.S. media company evaluated Tess as a licensed alternative to unlicensed AI art generation. Their legal team killed the deal because of unresolved copyright litigation around AI-generated works. This is the real bottleneck. Even when a company builds the "ethical" version, enterprise buyers can't touch it until courts clarify the legal landscape.
Kapwing has folded the technology into its main product as "Custom Kais," letting users emulate their personal style using reference images. The standalone royalty marketplace is gone.
The lesson here isn't that paying artists is wrong. It's that royalties address a moral argument, not a market one. The creative community's resistance to AI art runs deeper than compensation. And until copyright law catches up, the enterprise customers who might actually pay enough to make the economics work are stuck on the sidelines.