March 9, 2026
Royalties vs. Reality: Comment Chaos
We Tried Paying Artists Royalties on AI-Generated Work – Learnings
50% royalties, legal gambits, then a shutdown—commenters split between 'finally' and 'still theft'
TLDR: Tess.Design paid artists 50% for AI-generated images in their style, then shut down after two years. Commenters split between praising the transparency, slamming the legal foundation (base models trained on scraped images), and pitching a music-style royalty system—highlighting how hard it is to fairly pay creators for AI mimicry.
Tess.Design tried to do the impossible: pay artists when AI mimicked their style. They offered 50% royalties, advances to 25 artists, and a promise of “properly licensed” images. After two years, they shut it down—cue the comment section courtroom. Some cheered the transparency, with one reader giving “props for a postmortem” and likening it to publishing negative results, a rare move in startup land.
But the biggest sparks? Legality and ethics. Critics pounced on Tess fine-tuning a base model like Stable Diffusion—trained on the open internet—calling it a “veneer” of consent on top of questionable training data. Another camp pushed back on the whole premise: as one commenter put it, style isn’t protected by copyright, so why pay for it at all? Meanwhile, a pragmatic crowd floated a fix: an ASCAP/BMI-style system—those U.S. music royalty groups—for artists. One even joked about getting tiny music checks and wondered if illustrators were doomed to “coffee-receipt royalties.”
Numbers also fueled the drama: out of 325 cold emails to top-tier illustrators, 6.5% joined (about 21 models), and only 1 in 4 artists used their own AI model, which raised eyebrows about real-world usefulness. Love it or loathe it, the comments agree on one thing: paying artists for AI style is either the future—or a legal minefield with no map.
Key Points
- •Tess.Design launched in May 2024 as a marketplace for artist-specific, fine-tuned AI image models paying 50% royalties per use.
- •The platform used Stable Diffusion as a base, with models fine-tuned on consenting artists’ work and listed for subscribers.
- •Tess developed a legal framework with Fenwick asserting artists held copyright over stylistically transformed outputs for downstream licensing.
- •Artist outreach included 325 cold emails: ~50% response rate; 6.5% joined; 22.4% declined; 20.8% were maybes; 50.2% did not respond.
- •Tess shut down in January 2026 and reported lessons from seeding with 25 artists, advances of $300–$4,000, and limited artist adoption.