ai-video-pipeline

Deterministic video production scaffold for quick draft-to-export workflows.

Includes

  • SKILL.md - operational instructions and guardrails
  • scripts/init_project.sh - project folder + brief bootstrap
  • scripts/voiceover_natural.sh - natural-first VO auto-select (ElevenLabs -> OpenAI -> macOS fallback)
  • scripts/export_variants.sh - 16:9, 9:16, 1:1 renders
  • scripts/qa_report.sh - JSON output integrity summary

Quick start

# 1) Initialize project
PROJECT_DIR=$(./scripts/init_project.sh "CivicOS Launch Teaser" /tmp/video_projects)
echo "$PROJECT_DIR"

# 2) Put source video at:
# $PROJECT_DIR/src/master.mp4

# 3) (Optional) Generate natural voiceover from script text
# Put your script at: $PROJECT_DIR/assets/script.txt
./scripts/voiceover_natural.sh "$PROJECT_DIR/assets/script.txt" "$PROJECT_DIR/audio/voiceover.wav" narrator

# Optional script tags for expressive delivery:
# [[pause:500]] [[emph:critical phrase]] [[slow:this sentence lands here]]
# [[calm:steady and grounded]] [[urgent:we need action now]] [[inspiring:we can build this together]]

# 4) Export platform variants
./scripts/export_variants.sh "$PROJECT_DIR/src/master.mp4" "$PROJECT_DIR/exports"

# 5) Generate QA report
./scripts/qa_report.sh "$PROJECT_DIR/exports" "$PROJECT_DIR/qa/report.json"
cat "$PROJECT_DIR/qa/report.json"

Notes

  • Uses ffmpeg/ffprobe from PATH.
  • This is an MVP scaffold: expand with auto-script generation, voiceover, and caption styling next.