Graphix

A story goes in as natural language. What comes out is beats, images, layouts, compositions, print-ready pages. Somewhere between the input and the output, something interesting happens: the narrative gets decomposed into structural primitives that scaffold everything downstream.

That decomposition is what I'm actually studying. What is the fundamental representation of the kernels of storytelling? Not "how do you generate pretty pictures with AI," but: what are the minimal structural units that a story needs to become a visual sequence? Graphix is the first tool that binds everything required to explore that question in one place, exposed to an instrumented agent that lets me begin approaching it experimentally.

The product layer works: write story beats, the system scaffolds panels from narrative structure. Characters maintain visual consistency through IP-Adapter embeddings and LoRA associations. ControlNet handles pose and composition. A generation tree tracks every variant so you can branch and compare. 201 MCP tools give Claude control over the full pipeline from project creation through page layout and PDF export, with the ComfyUI MCP server handling the actual image generation.

Right now this handles static graphic novels: panels, pages, consistent characters, print-ready output. The next milestone is interactive panels with click-to-animate via image-to-video. After that, animated shorts: the full T2I to I2V to V2V pipeline.

A note on artists: this doesn't replace them. The structural decomposition I'm studying is exactly the kind of thing working illustrators and storyboard artists already do intuitively. The goal is to understand and formalize that intuition, not to automate it away. If anything, better formal representations of narrative structure make the case for why human editorial judgment matters more, not less: I'm hoping this project allows us to show not just that, but how, that which we value about creativity is, indeed, uniquely human.