🎬 22B Labs · Open Source Release

Short-Form Video: Plan First, Shoot Second

Shorts Engine — Open source CLI planning engine for YouTube Shorts, TikTok, and Instagram Reels

📅 April 2026 ✍️ 22B Labs ⏱ 5 min read

AI video generation tools are everywhere. Sora, Kling, Runway. But deciding what to make, in what order, to what platform spec, at what cost — that's still all manual work.

Shorts Engine automates the planning part. Feed it a request file and it outputs a complete production plan: platform specs, camera motion, B-roll recommendations, cost routing, execution order, recovery paths, and AI generation prompts. TypeScript CLI, 127 tests passing.

⚠️ What This Tool Does NOT Do

Shorts Engine does not generate actual video. AI video API integration (Sora, Kling, etc.), TTS generation, and platform auto-upload are not yet implemented. This is a planning engine — it tells you what to make and how, not make it for you.

"Philosophy → Scenario → Emotion → Technology.
Story comes before tools."

features

What Comes Out of One Request File

Output	Description
Validation + normalization	Input format check and standardization
Platform output spec	YouTube/TikTok/Reels specs auto-applied, out-of-range auto-corrected
Novel → shorts planning	Convert novel episodes to short-form scenarios
Motion planning	Camera movement patterns with repeat-prevention rules
B-roll planning	Contextually matched supplementary clip recommendations
Cost routing	5 deterministic rules auto-select cheapest backend
Execution plan	Per-node retry, fallback, and cost plan
Recovery simulation	Predicted auto-recovery paths on error
AI prompt generation	Ready-to-use prompts for video generation tools
Narrative DNA	studio philosophy-based scene_archetype + emotional_texture

platforms

Supported Platforms

▶️

YouTube Shorts

Rec. 30s / Min 15s / Max 60s
Search optimization, clear hook

🎵

TikTok

Rec. 20s / Min 10s / Max 45s
Motion energy, instant captions

📸

Instagram Reels

Rec. 20s / Min 10s / Max 45s
Visual consistency, brand finish

cost routing

5 Cost Routing Rules

Rule ACandidate score below 0.6 → no premium engine
Rule BCache hit → return immediately (always first priority)
Rule CBatch ≥5 + GPU available → prefer GPU
Rule DPremium engine only for final high-value stages
Rule ERetry cost > expected gain → go straight to fallback

usage

Getting Started

# Install
git clone https://github.com/sinmb79/Shorts-engine.git
cd Shorts-engine
npm install

# Run (human-readable output)
npm run engine -- run tests/fixtures/valid-low-cost-request.json

# JSON output
npm run engine -- run tests/fixtures/valid-low-cost-request.json --json

# Other commands
npm run engine -- prompt <file>    # Generate AI prompts
npm run engine -- analyze <file>   # Analysis report
npm run engine -- doctor           # System health check
    

request file

Request File Structure

{
  "intent": {
    "topic": "your video topic",
    "platform": "youtube_shorts",
    "duration_sec": 30
  },
  "constraints": {
    "budget_tier": "low",     // low / balanced / high
    "quality_tier": "balanced"
  },
  "style": {
    "hook_type": "curiosity",
    "pacing_profile": "fast_cut",
    "caption_style": "tiktok_viral"
  },
  "backend": {
    "preferred_engine": "local",
    "allow_fallback": true
  }
}
    

💡 Tests — npm test → 127 tests passed, 0 failed

output

Output Fields

validationRequest file format check result

platform_output_specPlatform specs + auto-corrections

motion_planCamera movement plan

broll_planB-roll clip recommendations

routingSelected backend + reason codes

execution_planProcessing order + per-stage cost

recovery_simulationError recovery paths

narrative_payloadStudio DNA contract structure

Get Started on GitHub

MIT License · Completely Free · TypeScript
127 tests passed · Node.js 24+
YouTube Shorts / TikTok / Instagram Reels

🔗 View on GitHub

22B Labs · The 4th Path GitHub: sinmb79 · the4thpath.com
The 4th Path: ⟨H⊕A⟩ ↦ Ω — Human × AI → a better world.