Style as a Service

Model Context Protocols (MCPs) are now everywhere. You use it to connect AI to Blender, Figma, your house, the neighbour’s dog and all of the sudden you’re chatting with your tools like friends who actually listen to you. They’re particularly interesting in that MCPs appear to be the sort of anti-silo, API-esque approach we need. MCPs act as this weird trifecta: buffer, gate, and helper all at once. This got me thinking. If we can gate access to functionality, could we not gate access to other less tangible things like style?

The Internet is basically free, but content (the text and images, the hard goods of the Internet) can be paywalled. Okay fine, and this is only true to varying degrees. But what about more ephemeral things like style? Style is this slippery thing that’s hard to pin down in words but fairly definable in tokens.

TollBit already lets publishers enforce their robots.txt policies and charge bots that want access to their content. Their two-sided marketplace connects publishers who want to license their work and AI companies that need to use it for their models. Instead of gating content from humans like a paywall, it gates automated bots and AI agents. Same concept, different target really.

But style is inherent to the AI discussion. We saw it a little bit ago with Greg Rutkowski, Polish fantasy artist, whose name appeared as a prompt on Stable Diffusion more times than Picasso or Da Vinci. People want dragons and Stability gave them a way to make them look like how they’re obviously supposed to, which is dramatically poised and lit. Style, the inherent and seemingly invisible choices of how to make something, could be a way for Rutkowski and billions of other creative people to actually make a buck or two with MCPs potentially.

We could license style as an MCP for creativity. Just as an MCP defines how a model accesses context, style licensing would define how an AI accesses and uses a creator’s signature style. Each request to generate in that style gets controlled, monitored, and monetised. An artist/musician/etc. uploads their work, a style-as-a-service platform has a style embedding stored behind a paywall, and AI models that want to generate in that style pay per request. For instance, a writer’s voice model might cost $0.005 per 1,000 tokens. An illustrator’s environment concept art style might run $0.02 per image. Or something like that. Titles.xyz is sort of doing this already, but is crypto-based. There is a model there somewhere but it’s going to take some work building the right gates, moats and redoubts for this that people want to use.

However, one thing we all know and love, fonts, already work sort of like this. Fonts are pure style (abstract shapes, not creative works like illustrations) yet their creators make money selling the right to use them. They’re licensed, prices scale with usage, and EULAs and compliance audits enforce them as much as they can or how much people feel guilty. This is where artists are with AI: they have a unique style which isn’t fully copyrightable, but it could absolutely be contractually licensed and restricted. The model exists. We just need to build this sort of marketplace before the frontier models finish hoovering up everything that isn’t nailed down.