A Guide to AI Novel Writing with NovelKit
A step-by-step guide to AI novel writing on NovelKit Studio — no complex prompt engineering required.
This AI novel writing guide is for authors new to NovelKit. You don't need to understand Hermes deeply — just the order of operations: DNA → characters & world → outline → write → review → sync. NovelKit runs the pipeline on studio.novelkit.cc; every piece of canon is stored as a file in the workspace.
Step 1: Sign in and create a novel
Go to studio.novelkit.cc and sign in with Google. Choose Quick Setup (a fast pilot) or Full Setup (a complete DNA for long-form fiction). Each novel is its own Hermes workspace — memory and canon are isolated.
Step 2: Fill out the Project DNA
Genre, tone, target reader, hook, style model (Đại Thần, if any). The DNA locks the rules for every agent — details in Project DNA. Getting the DNA wrong here means fixing hundreds of chapters later.
Step 3: Bootstrap characters & world
Run Character Architect and World Builder (in parallel). Output goes into database/characters and database/worldbuilding. This is the core of the story bible. Don't jump straight into writing a chapter.
Step 4: Outline and write chapter 1
Plot Weaver → chapter outline. Prose Writer → draft. Quality Auditor → review. PASS 85+ → sync commit. On a SOFT-FAIL: revise per the review notes. Pipeline: 9 steps.
Step 5: Scale to more chapters
plan_next picks up ready tasks; resume continues after an interruption; doctor checks health. The 5-layer memory updates after every sync — see the memory layer. See public output: Novel AI Gallery.
- How long do the first 3 chapters take?
- It depends on the DNA and review; a 3-chapter pilot usually takes a few hours to 1-2 days with human oversight.
Beginner mistakes and how to avoid them
- Skipping bootstrap — writing chapter 1 before characters/world exist as files leads to reworking 30+ chapters.
- A vague DNA — a genre like "xianxia + romance + system" with no locked ratio dilutes the voice.
- Bypassing review — manually syncing a chapter that hasn't passed contaminates memory.
- Ignoring review notes — the same SOFT-FAIL errors repeat.
After the first 10 chapters
Run doctor periodically. Check the Cultivation Clock / relationship graph. Update the master outline if an arc drifts. Style audit cron runs every 10 chapters (on the Hermes schedule). Only scale up in batches once the pilot's retention hits the platform's threshold.
Cluster: canon, memory, gate, long-form fiction.
FAQ: writing guide
- Do I need to know how to code?
- No — the Studio UI drives the pipeline.
- How soon do I get chapter 1?
- After bootstrap + outline, usually within the first working day.
- Can a human edit a chapter?
- Yes — in a sandbox, before re-review.
The full AI novel writing cluster: NovelKit blog.
End-to-end AI novel writing on NovelKit
An end-to-end AI novel writing guide on NovelKit starts at Studio (studio.novelkit.cc), not from a blank chat. The process: create a novel workspace → fill out PROJECT_DNA (genre, style_model, premise) → bootstrap the character + world database → master outline → write/review/sync pipeline chapter by chapter. The Lãng Khách orchestrator dispatches the genre squad through delegate_tool — you don't need to know how to code; the UI drives plan_next.
After bootstrap + outline, chapter 1 usually lands within the first working day — but the real success metric is chapter 10/30 retention, not chapter-1 speed.
The 9-step pipeline and beginner mistakes
- Skipping bootstrap — writing chapter 1 before a character/world file exists means reworking 30+ chapters.
- A vague DNA — multi-genre without a locked ratio triggers a DNA Depth SOFT-FAIL from the Auditor.
- Bypassing review — manually syncing without PASS ≥85 contaminates memory and breaks RAG.
- Ignoring review notes — the same SOFT-FAIL errors repeat and the breaker opens after 3 soft fails.
- Scaling too early — batching 20 chapters before a retention pilot drives up TCO.
After the first 10 chapters
Run doctor periodically. Check the Cultivation Clock / relationship graph. Update the master outline if an arc drifts for a valid canon reason. The style_audit cron runs every 10 chapters (schedule.json on Hermes). Only scale up once the Publisher/platform hits its retention threshold.
Canon, memory, and gate — the three pillars
File-first canon: PROJECT_DNA, database/*, outlines, chapters, reviews — disk beats chat. The 5-layer novelkit_memory only updates after a sync commit. The gate: PASS 85 / SOFT 70 / breaker 2/3/5. Resuming after an interruption never duplicates a sync.
Studio onboarding, step by step
Step 1: sign up at studio.novelkit.cc, create a novel workspace. Step 2: fill out PROJECT_DNA — genre, style_model, premise, cultivation speed if xianxia. Step 3: bootstrap the character + world database in parallel (Lãng Khách dispatches through delegate_tool). Step 4: Plot Weaver builds a master outline of 3-4 arcs. Step 5: plan_next → write → review → sync per chapter — the UI drives it, no coding required.
After 10 chapters: a scaling checklist
- Run doctor — health-check canon, index, memory
- Check the Cultivation Clock / relationship graph / plot_threads
- Review the score distribution — the ratio of SOFT vs. PASS
- Confirm the style_audit cron ran at chapter 10 (schedule.json on Hermes)
- Run a retention test before batching 20+ chapters — track the chapter 10/30 metric
NovelKit's 6 service lines: xianxia, romance, isekai, system, urban, sci-fi — each with its own genre squad. The novelkit-context plugin runs RAG + RRF within budget; novelkit_memory provides per-novel isolation for a multi-title catalog.
Blog cluster: canon, memory, gate, pipeline, What is NovelKit.
FAQ: AI writing for beginners
- Can I start from chat?
- Brainstorming is fine — canon authority has to live in the Studio workspace from bootstrap onward.
- Can I skip the outline and just write?
- Not recommended — you'll rework 30+ chapters without a master arc outline.
- When can I scale to 50 chapters?
- After a chapter 10/30 retention pilot, plus doctor and a style_audit at chapter 10.
The Studio UI drives plan_next — authors don't need to know the Hermes CLI. Lãng Khách dispatches the 6-service-line genre squad through delegate_tool; the gate — PASS 85 / SOFT 70 / breaker 2/3/5 — protects the catalog.
End to end: PROJECT_DNA → bootstrap the database → master outline → write/review/sync per chapter → periodic doctor → scale after a retention pilot. novelkit_memory only updates after a sync commit with PASS ≥85.
A guide to AI novel writing on NovelKit starts at Studio — not a blank chat.
After chapter 10: ongoing operations
Turn on weekly doctor monitoring, track review score trends, and run style_audit every 10 chapters. If the SOFT-FAIL rate rises, check the DNA or the arc outline — don't just polish the prose. Scale up in batches through plan_next once health is green.
Common mistakes for beginners
Skipping World Builder because you "want to write now" — the result: locations and factions contradict each other by chapter 8. Filling out the DNA carelessly — the agent doesn't know the tone, and output turns generic. Syncing on a SOFT-FAIL — contaminates memory, and fixing 20 chapters later costs more than fixing it immediately. Copy-pasting from ChatGPT into a chapter file — breaks provenance and gate consistency.
Quick Setup vs. Full Setup
Quick Setup is for a 3-5 chapter pilot to test the voice. Full Setup is required for 50+ chapter serials: a fully filled-out DNA, a worldbuilding guide pointer, and a style model (Đại Thần) if applicable. Publishers usually require Full Setup before signing off on a catalog.
Checklist before chapter 1
- PROJECT_DNA verified ✓
- ≥2 main characters have a profile in the database
- World rules summarized in worldbuilding
- Chapter 1 outline has active plot threads
- doctor health check green
Cluster: What is NovelKit, DNA, Pipeline, Gallery, Studio.