For fiction writers working with generative AI, the single greatest point of failure has always been narrative amnesia. Standard Large Language Model (LLM) chat interfaces forget eye colors after three prompts, invent conflicting backstory facts by chapter two, and wander off-plot whenever a scene demands complex choreography. Sudowrite addresses this architectural bottleneck through Story Bible—a persistent, interconnected relational database that functions as a single source of truth across a project.
Unlike a passive wiki (such as Notion, World Anvil, or Scrivener notes), Sudowrite’s Story Bible is dynamically wired into the generative pipeline. The metadata, constraints, and traits an author records in their Character cards, Worldbuilding lore, and Chapter Outlines actively condition how Sudowrite’s Draft engine, Write modes, and Chat assistant generate prose.
Understanding how to methodically structure these three pillars—Character Profiles, Worldbuilding Lore, and Chapter Outlines—before writing a single sentence of scene prose is what separates disjointed, generic AI text from coherent, voice-consistent long-form fiction.
The Upstream Foundation: Braindump and Synopsis
While characters, lore, and outlines represent the core architectural scaffolding, they do not exist in a vacuum. Story Bible operates as a directed flow where upstream fields inform downstream generations:
$$\text{Braindump} \longrightarrow \text{Genre & Style} \longrightarrow \text{Synopsis} \longrightarrow \begin{cases} \text{Characters} \ \text{Worldbuilding} \end{cases} \longrightarrow \text{Outline} \longrightarrow \text{Scenes} \longrightarrow \text{Draft Prose}$$
Before generating or organizing cards, authors either paste their existing planning notes into the Braindump (unstructured notes, dialogue fragments, premise pitches) or summarize their narrative trajectory in the Synopsis. If you click Generate All Characters or generate an outline automatically, Sudowrite queries the Synopsis (or defers back to Braindump if Synopsis is blank). However, serious authors rarely leave these generations unedited; they treat AI generations as drafts and manually calibrate their cards to establish precise narrative boundaries.
1. Structuring Character Profiles (The Characters Section)
The Characters section stores up to 2,000 individual character cards per project. Each card represents a discrete entity and includes both narrative attributes for the author’s reference and operational flags that instruct the AI on how to render voice and behavior.
Card Mechanics and Role Hierarchy
Every character is assigned a narrative tier using the role dropdown in the upper right of the card:
- Protagonist: Receives primary weight in point-of-view (POV) calculations, internal monologue prompts, and emotional stakes.
- Antagonist / Supporting / Minor: Establishes relational distance and contextual hierarchy during group scenes.
- Cards can be collapsed, reordered via drag-and-drop handles, and tracked across historical versions using the card-level clock icon.
The Standard Trait Schema
By default, each card comes with seven core trait fields designed to give the AI actionable boundaries:
- Pronouns: Fixes grammatical references to prevent pronoun switching during drafting.
- Groups: Maps the character to factions, crews, families, or organizations (e.g., The Silver Daggers, Crew of the Aegis). When a group name appears in a scene outline, the AI recognizes which cast members belong to that cohort.
- Other Names: Tracks aliases, titles, nicknames, and diminutives (e.g., Thomas, Tommy, The Black Fox). This field prevents the AI from treating an alias as a separate, newly invented person.
- Personality: Defines behavioral tendencies, flaws, temperament, and psychological impulses under stress.
- Background: Establishes canon backstory, origin wounds, and baseline history that the AI can draw on for internal thoughts and subtext.
- Physical Description: Anchors static visual markers—height, build, skin tone, eye color, scars, and signature attire—to eliminate physical continuity errors.
- Dialogue Style: Directs conversational cadences, verbal tics, syntax preferences, slang, formality level, and rhythm (e.g., speaks in short, clipped sentences; never uses contractions; dry sardonic cadence).
Custom Traits and Template Customization
Fiction genres often demand bespoke psychological or supernatural tracking. Authors can extend character cards in two ways:
- Card-Level Custom Trait: Clicking
+ Add Traitat the bottom of an individual card allows one-off additions, such as a Vampiric Bloodline Weakness or Hoverbike Model. - Global Schema Customization: Under the section header’s More menu (
•••), selecting Customize allows authors to modify default schemas for all character types. Authors frequently add fields for Enneagram / MBTI, Core Wound (The Ghost), Lie Believed, or Magical Affinity. Newly created fields can be applied across all existing cards in the project.
Ingestion Workflows: Smart Import vs. CSV
Authors transitioning existing manuscripts or series bibles into Sudowrite do not need to copy and paste cards manually:
- Smart Import: Accepts raw text or uploaded documents (
.txt,.docx,.rtf,.odt,.csv) up to 60,000 words. The AI analyzes the text, identifies up to 30 characters at a time, populates default trait fields, and provides an author validation modal before creating cards. Smart Import consumes zero generation credits. - Structured CSV Import: For comprehensive series rosters, authors can populate Sudowrite’s CSV template. Uploading structured CSV data bypasses AI interpretation entirely, creating unlimited cards instantly with columns mapping directly to card traits.
2. Architecting Worldbuilding Lore (The Worldbuilding Section)
The Worldbuilding section serves as the repository for environmental, mechanical, and cultural canon. Because fantasy, sci-fi, and historical fiction rely on strict ontological rules, Sudowrite organizes lore into specialized card templates rather than freeform text blocks.
Specialized Card Templates
Clicking + Add Element exposes pre-configured templates that prompt for genre-specific attributes:
- Setting: Geography, sensory atmosphere, light, weather patterns, social climate, and landmarks.
- Item: Appearance, origin, powers or functions, and the specific "Who owns this item" ownership field to prevent items from magically jumping inventories.
- Clue: Physical evidence, location found, and the vital "Implications" field (instructing the AI on what conclusions characters should or should not deduce upon examining it).
- Lore & History: Historical timelines, pivotal wars, treaties, legends, and forgotten knowledge.
- Faction / Organization: Internal hierarchies, uniforms, codes of conduct, public reputation, and rival groups.
- Magic System / Technology / Rules: Explicit cosmological laws, limitations, costs of magic, physical paradoxes, and failure conditions.
- Custom Elements: Fully author-defined categories configured through the
Customize Traitsmenu.
The Selective Saliency Rule: How the AI Accesses Lore
The most common mistake new authors make is assuming Sudowrite dumps every character and worldbuilding card into every prompt. Doing so would instantly overflow the context window, dilute instructions, and produce slow, erratic prose.
Instead, Sudowrite uses Selective Retrieval (Story Bible Saliency):
The Retrieval Rule: When generating Chapter Scenes and Draft Prose, the engine queries only the specific Character and Worldbuilding cards whose names or aliases appear explicitly in your active Outline or Scene instructions.
If your world contains 50 setting cards, but your Scene beat specifies The Sunken Archive, only the card for The Sunken Archive—alongside characters explicitly named in that beat—is retrieved and injected into the context window. This architectural constraint makes proper naming and alias management essential during the pre-drafting phase.
3. Engineering the Chapter Outline & Act Structure (The Outline Section)
The Outline section bridges high-level world lore and character profiles into a sequential macro-narrative. The Sudowrite Outline has no word limit, allowing it to house full multi-act structures spanning dozens of chapters.
Hierarchical Organization: Acts and Chapters
Authors organize their narrative roadmap using two organizational tiers:
- Acts (
+ Add Act): Act dividers act as structural waypoints (e.g., Act I: The Inciting Breach, Part II: The Crossing, or Midpoint Climax). These can be dragged across the outline to visualize narrative pacing. - Chapters (
+ Add Chapter): Discrete chapter units that contain narrative summaries of plot progressions, character developments, and pivotal reveals.
Generation, Rewriting, and Granular Control
Authors can generate the outline holistically or build it iteratively:
- Holistic Generation: Clicking Generate Outline synthesizes Genre, Synopsis, Characters, and Worldbuilding into an end-to-end multi-chapter sequence.
- Global Outline Rewrite: Under the Outline header’s More menu (
•••), authors can use Rewrite Outline with custom prompt instructions (e.g., “Increase tension in the second act by introducing a ticking clock regarding the orbital decay”). - Individual Chapter Rewrite: Clicking Rewrite directly within an individual chapter summary lets authors revise a specific beat (e.g., “Make Maya’s betrayal subtle rather than overt, ending on a cliffhanger”) without disturbing surrounding chapters.
The "Start Chapter" Linked Document Architecture
The Outline does not write prose directly inside the bible card. Instead, hovering over any chapter summary reveals the Start Chapter button (represented by an icon with a plus sign on a document).
Clicking this button generates a Linked Document in the project navigation bar:
- The newly spawned document is permanently tied to that chapter summary.
- The Draft tool inside that document inherits the chapter summary as its structural target.
- Clicking the chapter summary’s More menu (
•••) allows authors to inspect all linked documents associated with that plot point.
The Bidirectional Sync Loop ("Update")
In practice, drafting rarely follows an outline with 100% fidelity. Characters make unplanned choices, conversations veer into unexpected revelations, and scenes evolve.
Sudowrite accommodates discovery drafting via its Update mechanism:
- If your drafted chapter diverges from your pre-planned outline, you do not need to retype your outline manually.
- Clicking Update from the chapter summary’s More menu (
•••) prompts the AI to read the drafted prose in the linked document and automatically rewrite the Outline summary to match what was actually written. - This keeps downstream chapters aligned with canon without breaking narrative continuity.
4. Translating the Outline into Scene Beats Before Drafting
The critical handoff from pre-writing to prose generation occurs inside the Draft tool (the typewriter icon in the upper right of the text editor). This is where character profiles, worldbuilding lore, and chapter summaries crystallize into actionable generation prompts.
+-----------------------------------------------------------------------------------+
| SUDOWRITE DRAFT PANEL (Typewriter Icon) |
+-----------------------------------------------------------------------------------+
| 1. SCENE BEATS (Atomic Units of Action) |
| 1. Thomas arrives at [The Sunken Archive], shaking rain from his coat. |
| 2. Archivist Vane reveals the fractured seal; dialogue is clipped and tense. |
| 3. A noise echoes from the subterranean vault; Vane conceals the cipher. |
| |
| * Underlined terms indicate active Story Bible Detection (Cards linked to prompt) |
+-----------------------------------------------------------------------------------+
| 2. EXTRA INSTRUCTIONS |
| [ Third-person limited POV (Thomas). Gritty, sensory atmosphere. Fast pacing. ] |
+-----------------------------------------------------------------------------------+
| 3. ENGINE VALIDATION & EXECUTION |
| [Quick Fix Suggestions] --> [Word / Credit Estimate] --> [Generate Chapter] |
+-----------------------------------------------------------------------------------+
Deconstructing Scenes as "Atomic Units"
A chapter summary is too broad for an LLM to render as nuanced prose in a single pass. The Draft tool breaks the chapter down into Scenes: numbered, self-contained beats defined by a single location, timeframe, POV, or dramatic objective.
Authors can write these beats manually or click Generate Scenes to have Sudowrite convert the linked chapter summary into sequential scene beats.
Story Bible Detection Validation
As beats are populated, Sudowrite runs Story Bible Detection. Any character name, alias, setting, or worldbuilding item that matches an existing card is automatically underlined in the text box.
- If an entity is underlined, you have confirmed that the engine will load that card’s specific traits (dialogue style, physical description, sensory lore) into the LLM context window.
- If a name is not underlined, the author knows to either adjust the spelling to match the card or add the variant name to the character’s Other Names field.
Controlling Chapter Length via Beat Density
Sudowrite does not use an arbitrary "word count slider" to set draft length. The generated word count is directly governed by the number of scene beats provided:
- A chapter with 1–2 consolidated beats produces a concise, rapid scene (approx. 600–900 words).
- A chapter with 5–7 detailed beats generates an expansive, fully fleshed-out chapter (approx. 2,000–3,500+ words).
Fine-Tuning with Extra Instructions
Directly beneath the Scenes list lies the Extra Instructions field. Experienced authors use this field to enforce stylistic constraints that override or sharpen general settings:
- Specifying strict POV and tense (e.g., Third-person limited, past tense, deep close psychic distance through Thomas).
- Sensory mandates (e.g., Focus heavily on the smell of ozone and damp parchment; keep dialogue sparse).
- Structural targets (e.g., End the final beat on a sudden, unresolved sensory shock).
The Pre-Drafting Diagnostic Check
Before prose generation commences, clicking Generate triggers a structural diagnostic:
- Structural Warnings: The system flags scenes that are overly broad, cover too many locations at once, or pack conflicting character actions into a single beat.
- Quick Fix Suggestions: Authors can accept one-click recommendations to split overcrowded beats into discrete units.
- Resource Estimates: Displays estimated credit consumption and generated word count before execution, allowing writers to manage plan quotas effectively.
5. Author Best Practices: Avoiding the Three Pitfalls of AI Prose
To get publication-ready prose out of Story Bible, authors adhere to three foundational operating rules:
| Pitfall | Root Cause | Story Bible Solution |
|---|---|---|
| Vagueness Hallucinations | Leaving beats underspecified (e.g., "They discuss the artifact and leave"). | Write concrete physical actions into the scene beat ("Vane demonstrates the cipher box opening counter-clockwise, warning Thomas of the poison spring"). Specificity prevents the AI from inventing arbitrary lore. |
| Lore Contradictions | Outdated traits remaining on cards after major plot shifts. | Audit character motivation and background fields at the start of each Act. If a protagonist changes allegiance in Act II, update their card so the AI does not default to their Act I mindset. |
| Referential Ambiguity | Using unanchored pronouns or vague backstory references in beats ("She remembers what he did"). | Explicitly name entities and events in scene beats ("Elena recalls Marcus burning the census records in Chapter 4"). Draft only references linked documents and explicitly triggered cards. |
By treating Sudowrite’s Story Bible as an active structural harness rather than a static filing cabinet, fiction authors maintain ironclad narrative continuity across complex world systems, nuanced casts, and epic outlines—setting the stage for rapid, high-fidelity scene drafting.