Introduction
Overview of the Shotrunner autonomous video engine.
Shotrunner is a 10-stage autonomous video generation pipeline built for AI coding agents. It takes a series topic, fact-checks historical research, drafts high-retention narration, synthesizes studio-grade neural voiceover with phoneme lock, retrieves and filters visuals, and renders 1080×1920 broadcast-ready MP4s at −14.0 LUFS.
Overview
Shotrunner runs entirely on your local machine using CLI commands, gitignored local credential storage, and local FFmpeg/Whisper processing. Everything is deterministic, reproducible, and verifiable via automated gate commands.
Key Capabilities
- •Autonomous Research & Fact Checking: Queries multiple independent web sources to gather factual source material.
- •Retention Scripting: Generates 100–130 word scripts with hook verification, AI-cliché filters, and seamless loop closings.
- •Phonetic Pronunciation: CMU Arpabet mapping prevents mispronunciation of character names and proper nouns.
- •Multi-Source Visual Ingestion: Downloads high-resolution images with automated OCR watermark filtering and aspect-ratio validation.
- •Broadcast Audio Mastering: Mixes voice, ducked background music, and sound effects to −14.0 LUFS standards.
- •Deterministic State Ledger: Every stage logs progress to logs/series.log so interrupted runs resume without repeating passed stages.
Pipeline Architecture
Production runs across 10 stages. After initial research and scripting (Stages 00–02), the audio and visual branches execute in parallel before converging for final rendering and quality control (Stage 08):
