We Tried Paying Artists Royalties on AI-Generated Work – Learnings

Kapwing undertook an ambitious project called Tess.Design, an AI image generation platform launched in May 2024 with the unique goal of ensuring artists received 50% royalties whenever their fine-tuned style models were used. Positioned as the 'first properly-licensed' image generator, Tess sought to solve the widespread issue of AI tools training on unlicensed art and the associated copyright controversies. Despite a novel legal framework designed to establish clear copyright ownership for derivative works, the platform struggled to gain traction and was ultimately shut down in January 2026.

• The Problem: The AI art landscape was marred by controversy over unlicensed data training and legal uncertainty, leaving media companies hesitant to adopt AI. Tess aimed to provide a legally sound and ethically compliant alternative.
• Business Model: Artists submitted their work to fine-tune Stable Diffusion models, creating unique style models. Subscribers paid for access, and artists earned royalties from usage, with initial advances provided to founding artists.
• Artist Engagement: Out of 325 artists cold-emailed, only 6.5% agreed to join. Significant opposition stemmed from ideological anti-AI stances, fears of brand dilution, principled objections to AI art, and the risk of social backlash within the creative community. Those who joined were motivated by passive income, curiosity, and the desire to offload tedious tasks.
• Financial Performance: Tess generated just over $12,000 in gross revenue over 20 months, while incurring approximately $18,000 in artist advances and infrastructure costs. No artist earned additional royalties beyond their initial advance, highlighting the platform's inability to scale.
• Reasons for Failure: The shutdown was attributed to three main factors: unresolved legal arguments regarding AI copyright, which deterred enterprise adoption; poor timing, as the art community in 2024 was deeply hostile to AI; and the need for Kapwing to refocus resources on its core product.
• Learnings for Future Ventures: The author concludes that the underlying model of paying creators for their data or style isn't dead but is early. Key takeaways include the difficulty of a two-sided marketplace for fragmented creative industries, the importance of addressing brand dilution concerns, and the critical role of timing and focus.

The story serves as a cautionary tale and a detailed case study for anyone attempting to bridge the gap between generative AI and fair compensation for human creativity. It underscores the profound challenges, both ethical and practical, in building sustainable models in this rapidly evolving space, while still affirming the potential of such a vision for a more equitable future.