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
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."
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:
- Uploading to a public URL (tmpfiles.org / R2 / GitHub releases)
- Scheduling across platforms (AiToEarn MCP → YT + TikTok in one call)
- Handling failures (credential degradation, cold starts, per-platform validation)
- 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.
