Build an AI Short Video Generator

The Problem

Creating short-form video content at scale for YouTube Shorts, TikTok, and Instagram Reels is extremely labor-intensive. Manually scripting, recording narration, finding footage, adding subtitles, and uploading limits most creators to 1-2 videos per day — not enough to build traction on algorithm-driven platforms that reward volume and consistency.

The Solution

Build an automated pipeline that generates faceless short-form videos — from trending topic research through script writing, narration, stock footage selection, subtitle generation, assembly, and upload. Target: 50 videos/day at ~$0.05/video cost. No camera, no face, no editing skills needed — just AI + automation.

Architecture

Trending Topics API
       ↓
  AI Script Writer (Claude)
       ↓
  Text-to-Speech (ElevenLabs / OpenAI TTS)
       ↓
  Stock Footage Matcher (Pexels API)
       ↓
  Subtitle Generator (WhisperX)
       ↓
  Video Assembler (FFmpeg / MoviePy)
       ↓
  Upload Bot (YouTube API / TikTok)
       ↓
  Analytics Dashboard

Step-by-Step Walkthrough

1. Set Up the Project

mkdir video-generator && cd video-generator
python -m venv venv && source venv/bin/activate
pip install anthropic requests moviepy pydub whisperx srt schedule
sudo apt install ffmpeg

Environment variables needed:

export ANTHROPIC_API_KEY=sk-ant-...
export ELEVENLABS_API_KEY=...
export PEXELS_API_KEY=...
export YOUTUBE_CLIENT_SECRET=...  # for upload

2. Topic Research Engine

Find what's trending and likely to go viral:

# Research trending topics in your niche
# Use Google Trends API, Reddit trending, or curated topic lists
# The ai-video-generator skill has get_trending_topics() function

# Key niches with high RPM:
# - "AI tools nobody talks about" ($8-15 RPM)
# - "Psychology facts" ($5-10 RPM)
# - "Money/finance tips" ($10-20 RPM)
# - "Tech reviews" ($8-15 RPM)

3. Script Generation with Hooks

Every viral short follows this formula:

HOOK (0-3s): "Stop scrolling — this AI tool is replacing..."
BODY (3-50s): 3-5 punchy facts, each 1-2 sentences
CTA (50-60s): "Follow for more, link in bio"

Use the generate_script() function from the ai-video-generator skill. Key: the hook is everything — if they don't stop scrolling in 3 seconds, you lost them.

4. Voice Generation

Choose your voice strategy:

• ElevenLabs ($22/mo): Best quality, most natural, voice cloning
• OpenAI TTS (~$0.015/1K chars): Good quality, cheaper at scale
• Edge TTS (free): Decent quality, unlimited, good for testing

For a faceless channel, pick ONE consistent voice — it becomes your brand.

5. Stock Footage Matching

# For each script section, extract 2-3 keywords
# Search Pexels API for matching portrait (9:16) clips
# Download and cache clips for reuse
# Tip: build a local footage library to reduce API calls

6. Subtitle Generation

Subtitles are CRITICAL for shorts — 80% of viewers watch with sound off.

Style: bold white text with black outline, centered bottom third, 3-5 words at a time, word-by-word highlight for engagement.

7. Video Assembly

# Combine: background clips + narration + subtitles
# Output: 1080x1920 (9:16), H.264, AAC audio
# Add: slight zoom/pan on clips to avoid static frames
# Duration: 30-60 seconds optimal

8. Upload Automation

# YouTube Shorts: Use YouTube Data API v3
# - Upload via resumable upload
# - Set: #Shorts in title/description for Shorts shelf
# - Schedule uploads: 3-5 per day, spread across hours

# TikTok: Use unofficial API or Selenium automation
# - Upload via browser automation
# - Add trending sounds when relevant

9. Batch Production

# Daily batch: generate 50 videos overnight
# Store in output/YYYY-MM-DD/ directory
# Upload scheduler: drip 5-10 per day to avoid spam flags
# A/B test: same topic, different hooks → keep the winner

10. Analytics & Optimization

Track what works:

• Hook retention — if <30% watch past 3s, rewrite hooks
• Completion rate — target >60% for Shorts algorithm
• Click-through rate — thumbnail + title optimization
• Revenue per video — double down on high-RPM topics

Cost Breakdown

Revenue Timeline

Real-World Example

A creator launches a faceless "AI Tools Nobody Talks About" YouTube Shorts channel. The pipeline researches trending AI tools via Product Hunt and Hacker News, generates 60-second scripts with Claude highlighting one tool per video, narrates with a consistent ElevenLabs voice, overlays Pexels stock footage of people using laptops, and adds word-by-word highlighted subtitles. In month 1, they upload 500 videos and get 50K total views. By month 3, one video about an obscure AI image editor goes viral (2M views), pushing the channel past the YouTube Partner Program threshold. At month 6 with 3,000+ videos live, the channel earns $1,500/month from YouTube ad revenue ($12 RPM on tech content) plus $400/month from affiliate links to the AI tools featured in video descriptions.

Tips for Success

1. Consistency beats quality — post daily, the algorithm rewards frequency
2. First 3 seconds decide everything — spend 50% of effort on hooks
3. One niche per channel — don't mix finance and cooking
4. Recycle what works — same topic, different angle, new hook
5. Watch competitors — see what gets views in your niche, make it better
6. Build multiple channels — diversify risk and income sources

Related Skills

• ai-video-generator — core video generation pipeline
• ai-content-monetization — broader monetization strategies
• anthropic-sdk — Claude API for script generation