Novel Characters reads a novel or short story and produces, for each character, a profile, an image prompt that includes a full model-sheet layout, and a voice prompt. It does not generate images. The prompts are meant for a later image or speech step.
Long books are split into chunks, scanned by sub-agents, merged by name and alias, and then turned into character cards. When an outline file from the Novel Outline skill exists, the cast list is seeded from it. Output is JSON, Markdown and an interactive HTML report. Profile fields follow a chosen language, defaulting to Chinese, while image and voice prompts stay in English.
What You Get
- A cast.json with profiles, image prompts, voice prompts and sheet layouts
- A Markdown version and a double-click HTML report
- A story summary in the report language
- Resume support, since finished character cards are not redone
Common Use Cases
- Building a character bible from a novel before short-drama production
- Writing consistent image prompts for the same character across scenes
- Producing voice prompts for later speech generation
What Makes It Unique
It stops at prompts on purpose, because the image model, style and frame size are decisions for a later step. Quotes must match the source text word for word and image prompts may not contain character names, and both rules are enforced by a validator that was tuned on real model output. Any report language works, with the interface text translated on the spot for languages that are not built in.
Under The Hood
The skill is an instruction file with Node.js scripts that use only the standard library. It runs shell commands to chunk the book, merge rosters, assemble cards, validate and render, and it starts sub-agents to scan chunks and write cards in parallel when the environment supports it. It reads the novel text file and writes JSON, Markdown and HTML files plus working files into a work folder. It does not fetch web pages or call outside APIs and needs no API key.