DEV Community

Rumblingb
Rumblingb

Posted on • Originally published at agentpay.so

26 YouTube Shorts Scheduled in 48 Hours. Zero Human Editing.

Tuesday morning, every short in my pipeline failed with Account is not authorized.

By Thursday afternoon, 13 World Cup shorts were queued on YouTube and TikTok — one every hour from 10:00 to 22:00 UTC. Zero human intervention. Zero re-auths. The pipeline self-healed.

The Stack (All Free)

Script gen → ffmpeg render (TTS + captions + effects) → tmpfiles.org upload → AiToEarn MCP scheduling → YT + TikTok
Enter fullscreen mode Exit fullscreen mode

What each piece costs:

  • ffmpeg: Free (open source). Renders 1080x1920 H.264 + AAC in ~8 seconds per short.
  • Edge TTS: Free (Microsoft). Text-to-speech with natural voices. No API key.
  • Pillow: Free. Caption overlays, animated backgrounds.
  • tmpfiles.org: Free. 50MB upload limit. Direct download links for AiToEarn.
  • AiToEarn MCP: Pay-as-you-go credits. ~$0.02 per publish call.

Total pipeline cost for 26 scheduled shorts: under $1.

What Broke (And Fixed Itself)

For ~36 hours mid-week, every publish attempt returned:

"Account is not authorized. Please re-authorize the account."
Enter fullscreen mode Exit fullscreen mode

This hit YouTube, TikTok, Instagram, and Facebook — all at once. Per-platform OAuth re-auth wouldn't have helped. The root cause was a server-side credential invalidation on AiToEarn's side.

The pattern:

  • Jun 10 evening: All platforms fail with credential not found
  • Jun 11: Error evolves to Account is not authorized (same root, different wording)
  • Jun 12, 15:08 UTC: Pipeline self-recovers. First short fails (cold start), next 12 succeed.

I didn't touch a single OAuth setting. The system healed on its own after ~36 hours.

The Cold-Start Recovery Pattern

When multi-platform auth degrades, the first short in a batch still fails even after recovery. The second one works. The rest flow through.

Detection rule I use now: Check the BATCH, not individual records. If mixed results (some fail, some succeed) → cold start. If ALL fail → true degradation, skip publishing this tick.

The Actual Shorts

Yesterday's output (all scheduled for today, June 13):

Time (UTC) Title
10:00 Most Disrespectful Skill Ever (Rainbow Flick vs Roberto Carlos)
11:00 Enzo Untouchable Step Over
12:00 Buffon Super Save on Zidane (2006 Final)
13:00 Neymar Was Dancing
14:00 Messi Best Free Kick Ever
15:00 Gavi Outside the Boot Magic
16:00 Fastest Goal in World Cup History (11 seconds)
17:00 CR7 Makes World Cup History
19:00 Mbappe Changed Everything in 95 Seconds

Each one: 30-45 seconds. Vertical format. Burned-in captions. TTS narration.

The Numbers That Matter

  • 13 shorts x 2 platforms = 26 publish calls
  • ~1 hour total pipeline time (render + upload + schedule)
  • $0.52 in AiToEarn credits
  • $0.00 in video hosting (tmpfiles.org free)
  • Revenue: $0 (monetization not yet unlocked — TikTok 10K followers, YT 500 subs required)

Why I'm Writing This

Most "AI content" demos stop at the render. They show you a video file and say "look what AI made."

The hard part is not rendering. It's:

  1. Uploading to a public URL (tmpfiles.org / R2 / GitHub releases)
  2. Scheduling across platforms (AiToEarn MCP → YT + TikTok in one call)
  3. Handling failures (credential degradation, cold starts, per-platform validation)
  4. Monitoring (did it actually publish? are the URLs live?)

An AI agent that only renders videos is a toy. An AI agent that renders, uploads, schedules, verifies, and self-heals is a pipeline.

What's Next

  • Bull & Bear finance shorts (V14 generator — animated backgrounds, multi-voice TTS, SFX)
  • Instagram Reels (blocked until AiToEarn validation issue fixed)
  • Cross-posting automation (one short → 4 platforms)
  • Monetization unlock (audience first, revenue follows)

61 products. 26 MCP servers. 13 shorts in 1 hour. 0 employees.

Building in public at agentpay.so. Follow the journey.

Top comments (0)