Short fiction

For standalone stories of roughly 5,000–25,000 Chinese characters across 3–8 chapters. Entry: Short fiction → Library → New short fiction (短篇 → 作品库 → 新建短篇). The current engineering loop includes creation, review, targeted rewriting, versions, and export.

Preparation

Prepare a one-sentence premise, target readers, length, core conflict, and intended ending effect. A separate world is not required.

Complete workflow

1. Creative intent: define goals, length, and constraints.
2. Story design: establish choices, conflict escalation, and ending; confirm AI candidates.
3. Chapter cards: check each chapter's role and continuity; avoid dividing a short story into repetitive episodes.
4. Prose: generate, edit, and adopt chapter by chapter; inspect the prose actually saved.
5. Whole-story review: examine issues with chapter evidence and select parts for targeted rewriting.
6. Versions and export: after completion checks pass, keep a frozen version and export Markdown, TXT, or JSON.

Ready to deliver?

Check chapter completeness, actual versus target length, pending candidates, and serious issues. After prose changes, an old review may no longer apply and needs rechecking.

Publishing preserves a local frozen version; it does not automatically post the work to a public website.

Revision and derivation

Targeted rewriting fixes specific chapters while retaining confirmed content elsewhere. Expansion and derivation (扩展与派生) can explicitly derive a world; later world changes do not automatically rewrite the original story.

Common questions

Actual length differs from the target? Check completion results and chapter lengths, adjust chapter tasks, then expand or compress selectively. A target is not a guarantee of precise model counting.

Generation finished, but publishing is blocked? Check missing chapters, unadopted candidates, stale reviews, and unresolved serious issues; handle each as prompted.

Continue: AI generation and adoption · Export and delivery.