This one is noteworthy especially in terms of technique. As I mentioned in a previous post, Anthropic’s Claude LLM recently expanded its context window, to 100,000 tokens, or what I’ve read amounts to approximately 75K words. This was roughly the length of my first (non-AI) novel, The Lost Direction.
Claude has been my preferred tool for generating the flash fiction slice of life pieces which compose the last 8 or 10 books in this series, so I thought I would give it a try.
I don’t know if the tool was just having a bad day when I used it, or if it’s just slow af, but it took quite a while for Claude to process the text. And it seemed that whenever I did a query/prompt against it in the same conversation, it would then choke on it, having to go back and apparently parse the whole thing again.
It was not an ideal experience hitting those processing wait times again and again. But the results speak for themselves, especially for fans of The Lost Direction, and Quatriana in general. I had Claude generate a bunch of story suggestions based on the text – side stories based on lore mentioned in the main book. And it performed that task admirably, as well as the task of actually fleshing out the chosen stories.
Some of what it generated just made no sense relative to the details I input via my novel dump. In one notable case, it mixed up attributes of two main characters, resulting in unusable tales. In a few cases, they were just lackluster for the characters and topics represented, etc. But by and large, it did a pretty good job, and in some cases I think some of these tales probably rise to the level of “canonical” Quatria lore for my tastes (like the one about the training of a young augur, for instance). But the rest are absolutely good quasi-canon & apocrypha that once again does a great job fleshing out this massive legendarium that I am building.
Here’s some of the art, all done in Midjourney v5.1: