● Independent Creator Prototype • Built on Claude API

Narrative Continuity & Shot Generation for AI Micro-Dramas

EpisodeLoom helps independent creators and small studios break literary screenplays into verified shot-by-shot visual prompts while maintaining long-horizon character continuity across serialized episodic stories.

View Pipeline Sample Request Early Updates

From Literary Script to Structured Payloads

A real sample showing how episodic scenes decompose into deterministic JSON ready for video generation engines.

scene_to_diffusion_pipeline.json Claude Sonnet 5.5 Output
INPUT: EPISODE 03 / SCENE 02 
// Story Bible constraints active:
// Character: "Minh" (Detective, 34yo, worn brown trench coat, silver signet ring)
// Location: "Old Saigon Alley, neon drizzle, 9:16 vertical ratio"

[SCENE BEAT]
Minh stops under the flickering red neon sign of the tea house.
Rain trickles down his brimmed collar. He glances sharply over his
shoulder as footsteps splash through puddles behind him.
He slips his hand inside his coat, gripping the brass tape recorder.
OUTPUT: STRUCTURED SHOT PAYLOAD 
{
  "scene_id": "EP03_SC02",
  "continuity_verified": true,
  "shots": [
    {
      "shot_index": 1,
      "duration_sec": 3.0,
      "camera": "Medium close-up, low angle, slow push-in",
      "lighting": "Moody cyberpunk rain, red neon rim light",
      "diffusion_prompt": "Cinematic 9:16 vertical, 34yo Asian detective Minh in wet brown trench coat under buzzing neon sign, water droplets on wool, subtle head turn, 35mm film grain --ar 9:16",
      "target_engine": "Kling_v2 / ComfyUI_Wan2.1"
    }
  ]
}

Tiered Model Architecture

Pairing deep reasoning with high-throughput structured extraction.

01. STORY BIBLE Narrative Director Claude Opus 5.5 Reasoning 02. SHOT ENGINE Scene Decomposition Claude Sonnet 5.5 JSON Schema 03. RENDER DISPATCH Automated Batches Diffusion Engines & ComfyUI
OPUS 5.5 COGNITIVE CORE
1. Series Bible & Lore
Maintains cross-episode narrative arcs, emotional progression, and physical costume tags across 40–80 episodes without contradiction.
SONNET 5.5 FAST ENGINE
2. Shot-by-Shot Prompting
Breaks 2.5-minute literary screenplays into exact 3-second visual camera angles, camera motion cues, and atmospheric lighting directives.
DETERMINISTIC VALIDATION
3. Tool-Use & JSON Schema
Enforces strict JSON schema validation, eliminating prompt parser failures when dispatching batches to video generation clusters.
Transparent Project Status: EpisodeLoom is a bootstrapped, independent pre-launch prototype started in 2026. Core prompt chaining, continuity validation, and schema decomposition logic are currently tested on pilot scripts by the founder. We are not yet open for public accounts.
DN

Duong Nguyen

Founder & Builder • Ho Chi Minh City, Vietnam

System builder and AI media practitioner exploring serialized storytelling workflows. Building EpisodeLoom out of firsthand frustration with character drift and manual prompt fragmentation when producing episodic vertical micro-dramas with AI diffusion tools.

contact@episodeloom.com