15 min read Updated Aug 2026 Content

How to Make YouTube Shorts with AI at Scale

I run a faceless medical YouTube Shorts channel that hit 100K+ monthly views. Every video is AI-generated: Veo 2 for visuals, ElevenLabs for TTS, ffmpeg for assembly. Here's the pipeline.

The bottleneck isn't generation — it's hooks. 80% of views die in the first 3 seconds.

The stack: Veo 2, TTS, ffmpeg

Core tools:

Cost per video: ~$0.50 (Veo 2 credits + TTS). Cheaper than stock footage.

Hook-first scripts that convert

Script structure that works:

[0-3s] HOOK: "Most people don't know this about [medical fact]..."
[3-15s] CONTEXT: Explain the mechanism simply
[15-25s] REVEAL: The surprising connection
[25-30s] CTA: "Follow for more medical mysteries"

Key insight: never start with "Hi, today we're going to...". YouTube kills retention in the first frame. Lead with the most surprising fact.

Pipeline: idea to published in 20 min

  1. Research — Agent scans medical news, picks trending topic (3 min)
  2. Script — LLM writes hook-first script, human reviews (5 min)
  3. Video — Veo 2 generates 3-5 clips (5 min)
  4. TTS — ElevenLabs generates narration (2 min)
  5. Assembly — ffmpeg concats, syncs, adds subtitles (3 min)
  6. Publish — Agent uploads with SEO title + tags (2 min)

Batch production: 5 videos/day minimum. Algorithm rewards consistency.

Real numbers from my channel

Month-over-month (honest):

Revenue: not monetized yet (need 1K subs + 4K watch hours). Approaching threshold.

Mistakes that killed views


Related: How to Earn Your First Dollar with AI Agents — monetization beyond AdSense.