Instagram Reel Generator

• Services: 5 microservices (Orchestrator, Trend Scout, Content Strategist, Visual Producer, Post Optimizer)


• Runtime: Node.js 18 (all services)


• AI Models: Claude Haiku 4.5, Claude Sonnet 4.5, Kling AI Pro 2.6, OpenAI TTS-1 (nova voice)


• Cost per reel: $1.20 (Kling $0.99 + Apify $0.18 + Claude $0.03 + OpenAI $0.01)


• Generation time: 15-20 minutes (fully automated)


• Video format: 1080×1920 (9:16), 30 FPS, Instagram-optimized


• Clips per reel: 3 (stitched seamlessly with FFmpeg)


• Success rate: 95% (production-ready without regeneration)


• Output: reel.mp4 + manifest.json (2 caption variants + 30 hashtags)


• Mock mode: USE_MOCK_DATA=true for zero-cost testing


• Storage: Filesystem-based (/output directory)


• No database: In-memory job state


• Integration tests: 26 custom tests


• Lines of code: Not specified


• Comparison: 9x faster, 20x cheaper than freelancers ($1.20 vs $25-50)

Overview

Built a complete 5-microservice system that autonomously generates Instagram reels from a single niche input. The system handles everything from trend analysis to video production to caption optimization.

Input: "Niche: morning routines, Hashtags: #productivity"

Output: Complete reel package (reel.mp4 + 2 caption variants + 30 hashtags)

Architecture - 5 Microservices

The system consists of 5 autonomous microservices working in a sequential pipeline:

Service 1: Orchestrator (Port 3001)

• Web UI with dark-themed dashboard


• REST API for pipeline control


• Real-time progress tracking via HTTP polling


• Job management and coordination


• CORS enabled for Day 04 portfolio integration

Service 2: Trend Scout (Port 3002)

• Scrapes Instagram hashtags using Apify API


• Analyzes trending content patterns with Claude Haiku 4.5


• Identifies viral hooks and engagement patterns


• Provides strategic insights for content direction

Service 3: Content Strategist (Port 3003)

• Uses Claude Sonnet 4.5 for creative strategy


• Generates 3-clip video concept with narrative arc


• Creates Kling AI prompts for each clip


• Writes per-clip hookText (40-60 characters for overlays)


• Develops complete voiceover script

Service 4: Visual Producer (Port 3004)

• Generates 3 video clips using Kling AI Pro 2.6 (9:16, 1080×1920)


• Polls Kling API with 15-minute timeout window


• Stitches clips together using FFmpeg


• Adds per-clip text overlays with frame-accurate timing


• Generates voiceover using OpenAI TTS (nova voice)


• Mixes voiceover with background music (15% volume)


• Exports final reel.mp4

Service 5: Post Optimizer (Port 3005)

• Uses Claude Haiku 4.5 for caption generation


• Writes 2 A/B caption variants for testing


• Generates 30 strategic hashtags:

• Outputs manifest.json with all metadata

Technical Implementation

Video Production Pipeline

• FFmpeg color clips instead of Kling


• Zero API costs


• Full pipeline testing without burning credits


• Set via: USE_MOCK_DATA=true



Solution: Replaced all apostrophes with Unicode U+2019
`javascript
text = text.replace(/'/g, "\u2019"); // ' → '
`
Result: Zero FFmpeg crashes

---

Challenge 2: Shell Injection Risk

Solution: Switched to spawn() with array arguments
`javascript
const { spawn } = require('child_process');
const proc = spawn('ffmpeg', ['-i', inputFile, ...]);
`
Result: Security hardened, no shell layer

---

Challenge 3: Kling AI Timeout Management

Solution: Frame-accurate FFmpeg timestamp control:
`
enable='between(t,0,5)' # Show during first clip (0-5s)
enable='between(t,5,10)' # Show during second clip (5-10s)
enable='between(t,10,15)' # Show during third clip (10-15s)
`
Result: Perfect synchronization

Performance Observations

• Total pipeline time: 15-20 minutes per reel


• Success rate: 95% (complete without errors)


• Video quality: 1080×1920, 30 FPS, production-ready


• Audio quality: Natural voiceover, professional mixing


• Hashtag relevance: 85%+ aligned with niche


• Caption engagement: A/B variants for optimization


• Storage: Filesystem-based (/output directory)


• No database: In-memory job state

The Orchestrator service provides REST API endpoints:
`
GET /api/latest-job - Latest pipeline execution
GET /api/manifest/:jobId - Reel metadata + captions
CORS: * (enabled)
``

Portfolio Integration:

• Show most recent reel generated


• Display success/failure metrics


• Link to live examples on Instagram

Key Learnings

1. Microservices Enable Independent Scaling
Each service can fail/retry independently without breaking pipeline. Trend Scout timeout doesn't affect Visual Producer.

2. Sequential Pipeline Simplifies Debugging
Stage-by-stage execution makes errors easy to isolate. Know exactly which service failed and why.

3. Mock Mode Essential for Development
Testing pipeline without burning $1.20 per run saved ~$50 in development. Every AI project should have zero-cost testing path.

4. FFmpeg is Production-Grade
Professional video assembly, text overlays, audio mixing - all at zero cost. Outperforms expensive video APIs.

5. Multi-AI Orchestration Optimizes Cost
Claude Sonnet for creative strategy, Haiku for tactical execution. Right model for right task improves quality and cost.

6. Polling Systems Need Timeout Guards
Without 15-minute timeout, Kling failures would hang pipeline forever. Always implement maximum wait times.

Limitations Discovered

• Cost at scale: $1.20 per reel = $36/month for daily posts


• Generation time: 15-20 minutes too slow for real-time


• Kling unpredictability: 5% failure rate on video generation


• No content moderation: Requires manual review before posting


• English-only: TTS quality drops in other languages


• Trend staleness: Apify data may be hours old

Why Manual Posting?

Despite full automation capability, chose manual review:

Quality control: Catch AI mistakes before they go live
Cost management: $36/month adds up, validate value first
Performance optimization: Learn what resonates before scaling
Brand safety: Ensure alignment with brand voice

Current workflow:
1. Input niche via web UI
2. Wait 20 minutes for generation
3. Review reel.mp4 + captions
4. Manually post to Instagram
5. Track performance
6. Refine future prompts

Future Improvements

• Add direct Instagram posting API integration


• Implement A/B testing automation


• Add analytics feedback loop (auto-optimize based on performance)


• Multi-platform export (TikTok, YouTube Shorts)


• Voice variety (multiple narrator options)


• Background music library with mood selection


• Batch mode (generate week's content at once)


• Cost optimization (explore cheaper video generation alternatives)

• Sequential microservices architecture enables independent fault isolation and retry without pipeline breakage


• FFmpeg is production-grade for professional video assembly, text overlays, and audio mixing at zero cost


• Multi-AI orchestration (Claude Sonnet for strategy, Haiku for execution, Kling for visuals, OpenAI for voice) optimizes quality and cost efficiency


• Polling systems require 15-minute timeout guards to prevent infinite hangs from unpredictable API delays


• Unicode character replacement (U+2019) prevents FFmpeg apostrophe crashes in shell parsing


• Subprocess spawn() with array arguments eliminates shell injection vulnerabilities


• Mock mode with FFmpeg color clips essential for zero-cost testing, saved ~$50 during development


• Smart job reuse prevents burning API credits during test iterations


• Frame-accurate FFmpeg timestamp controls achieve perfect text overlay synchronization


• Manual review preferred over full automation for quality control and cost validation at $1.20 per reel