Mucky Master Sheet Creator The Mucky Master Sheet Creator, or MMC for short, was born from a simple problem: A book can be damn long. And at some point, it becomes pretty stupid to run the entire work through some language model for every new task and say: "Here. Read this again. And now please tell me who is actually who." So I decided to solve the problem properly once and for all. The MMC takes a complete book or chapter material apart and distributes the analysis across many small AI jobs. In the process, it collects things like: - Characters - Traits - Roles - Relationships - Locations - Objects - Recurring connections - Everything else that might become important later
And it doesn't work on the principle of: "Something was mentioned once, so now it will be the truth forever." But rather: first collect, then merge, then condense, then check again if all that stuff even fits together.
Many small jobs instead of one gigantic request For me, this is the far more sensible way. If it's mentioned somewhere in chapter 2 that a character has blonde hair, a scar appears in chapter 17, and in chapter 31 a specific vehicle suddenly belongs to them, then I want exactly this information merged later. Not some pretty-sounding AI fantasy. That's why the MMC deliberately builds a pipeline. Roughly: Intake → find characters → extract information → condense → merge book knowledge → Entity-Sweep → Consumer-Package
That sounds more complicated than it is in daily use. Basically, it's just a hardworking digital idiot who reads everything, writes index cards, and afterwards at least knows where he put them.
And yes: This costs computing power. But for me, that problem is pretty limited. I can run several AI models locally at the same time. So I just distribute the work across several workers and let the machines toil. That costs electricity and a bit of time. I can live with that. Anyone who thinks they have to send every single analysis job through some cloud service like Claude can, of course, feel free to do so. I would just lay out my wallet beforehand as a precaution. Then at least it knows where the pain is coming from.
Not the ultimate truth The MMC isn't supposed to pretend it's omniscient. It doesn't build a final, holy character sheet. It first collects the material for it. That's important to me. The machine is allowed to search, sort, and condense. I still make the important decisions myself afterwards. So the basic principle is quite simple: Machine collects. Machine hauls. Human decides.
What's the point of all this? Because other programs have significantly less work afterwards. The Mucky Book Illustrator can access the collected knowledge and depict characters more consistently. Speakers can be assigned more cleanly. And above all, the entire book doesn't have to be reread for every new production step. A thick, cumbersome work is thus turned into a compact, machine-readable memory.
And that's exactly what the MMC is for me: not a writing program, not an illustrator, and not a chatbot. More like a very hardworking librarian with several brains at once. And unlike a real librarian, he doesn't complain when I slam a whole book on his desk at three in the morning.
Gallery
