DOCS/CORE ARCHITECTURE/Pipeline & Architecture

Pipeline & Architecture

High-level overview of the 10-stage pipeline, parallel branches, and quality gates.

Shotrunner executes 10 discrete stages with dual parallel branches to maximize throughput while preventing wasted compute.

Parallel Execution Model

After Stage 02 completes and writes final.md, execution forks into two independent parallel branches:

•Audio Branch (03 → 04): Synthesizes neural voiceover (audio.wav) and computes frame-accurate Whisper subtitle timings (subtitles.srt).
•Visual Branch (05 → 06 → 07): Computes scene demand (needs.md), downloads imagery with OCR watermark rejection, and sorts frames into the series library (images/).

Both branches read only immutable inputs (final.md and config.yaml). When both report completion, Stage 08 renders the final video.

10-Stage Pipeline Matrix

STAGE	PHASE	OUTPUT	VALIDATION / GATE
00. Series Scaffold	Sequential	series/<name>/ & config.yaml	shorts series --name <name>
01. Live Research	Sequential	research/*.md & raw.md	shorts verify-raw (≥2 domains)
02. Script Transform	Sequential	final.md + phonemes	shorts verify-script (100–130w)
03. TTS Narration	Audio Branch	audio.wav	shorts tts (fish \| openai \| elevenlabs \| google \| azure \| polly)
04. Whisper Subtitles	Audio Branch	subtitles.srt & words.json	shorts subtitles (frame-sync)
05. Shot Planning	Visual Branch	needs.md	shorts verify-demand
06. Image Scraping	Visual Branch	raw-images/*	shorts download (OCR filter)
07. Vision Sorting	Visual Branch	images/*	shorts verify-sort (shot tags)
08. Render & QC	Convergence	short.mp4	shorts lint → render → qc (−14 LUFS)
09. YouTube Upload	Publishing	Live YouTube Short	User-Permission Gated

Quality Gates & State Ledger

Every stage records a PASS token to series/<series>/logs/<slug>.log. If any gate fails, execution halts immediately with actionable diagnostics. On resume, completed stages are skipped automatically.

Was this "Pipeline & Architecture" page helpful?

Secrets & API Setup

Configure Fish Audio keys, Google Cloud OAuth, and YouTube channel tokens.

Series Configuration

Series settings, voice profiles, subtitle styling, and custom channel mappings.