Cognotik | NarrativeGenerationTask

ExecutionConfig.json JSON

{
  "subject": "The Last Martian",
  "target_word_count": 5000,
  "number_of_acts": 3,
  "scenes_per_act": 3,
  "writing_style": "cinematic thriller",
  "tone": "suspenseful",
  "point_of_view": "third person limited",
  "generate_scene_images": true,
  "generate_cover_image": true,
  "narrative_elements": {
    "theme": "isolation",
    "conflict": "man vs nature"
  }
}

→

SessionTask UI Rendered

Overview

Outline

Act 1 Scene 1

Images

Scene 1: The Red Dust

Setting: Ares Base Alpha | Characters: Elias Thorne

The wind didn't howl on Mars; it hissed. Elias Thorne pressed his gloved hand against the reinforced polycarbonate of the hab-dome...

[Generated Scene Illustration]

Test Workspace: NarrativeGenerationTask

Explore the generated artifacts, including narrative_data.json, scene illustrations, and character reference portraits.

Configuration Parameters

Field	Type	Description
`subject`*	String	The core subject or scenario to develop into a narrative.
`target_word_count`	Int	Desired length (Default: 5000).
`number_of_acts`	Int	Structural acts, typically 3 or 5 (Default: 3).
`scenes_per_act`	Int	Average scenes per act (Default: 3).
`writing_style`	String	e.g., 'literary', 'thriller', 'technical' (Default: 'literary').
`tone`	String	e.g., 'dramatic', 'humorous', 'suspenseful' (Default: 'dramatic').
`point_of_view`	String	e.g., 'first person', 'third person limited'.
`detailed_descriptions`	Boolean	Include vivid sensory details (Default: true).
`include_dialogue`	Boolean	Include character dialogue (Default: true).
`generate_scene_images`	Boolean	If true, uses DALL-E/Midjourney to visualize each scene.
`generate_cover_image`	Boolean	Generates a visual seed for the entire narrative.
`revision_passes`	Int	Number of self-editing passes per scene (Default: 2).

Task Execution Lifecycle

Phase 1: Narrative Analysis

The agent performs a deep reasoning pass to establish themes, character arcs, and world-building constraints based on the subject and input files.

Phase 2: Multi-Level Outlining

Generates a high-level act structure, then expands each act into detailed scene summaries including purpose, emotional arcs, and character states.

Phase 3: Iterative Generation

Writes scenes one by one. Each scene receives the full context of previous "Key Moments" and "Character States" to ensure continuity. Optional revision passes improve prose quality.

Phase 4: Final Assembly & Data Export

Compiles the full narrative into a single document and exports all metadata, including character arcs and scene statistics, to narrative_data.json.

Embedded Execution

Invoke the Narrative engine programmatically using the UnifiedHarness.

import com.simiacryptus.cognotik.plan.tools.writing.NarrativeGenerationTask
import com.simiacryptus.cognotik.plan.tools.writing.NarrativeGenerationTask.NarrativeGenerationTaskExecutionConfigData
import com.simiacryptus.cognotik.plan.tools.writing.NarrativeGenerationTask.Companion.NarrativeGeneration

// 1. Define the Job
val executionConfig = NarrativeGenerationTaskExecutionConfigData(
    subject = "The Silicon Uprising",
    target_word_count = 10000,
    writing_style = "cyberpunk",
    tone = "noir",
    generate_cover_image = true,
    generate_scene_images = true,
    revision_passes = 2
)

// 2. Run via Harness (Headless/Serverless mode)
harness.runTask(
    taskType = NarrativeGeneration,
    typeConfig = TaskTypeConfig(), // Default static config
    executionConfig = executionConfig,
    workspace = File("./output/stories"),
    autoFix = true
)

💡 Pro Tip: Visual Consistency

When generate_cover_image is enabled, the engine uses the generated cover as a visual seed for all subsequent character and scene illustrations, ensuring stylistic coherence across the entire narrative.