Web Application
Short Publisher
Autonomous daily YouTube Shorts factory orchestrating script, voice, and video AI end-to-end.
- TypeScript · Python
Overview
An end-to-end pipeline that produces and publishes a cinematic ~60-second YouTube Short every day with zero manual steps. A GitHub Actions cron runs an 8-step Python orchestrator: pull the next topic from NeonDB, write the script with Claude Sonnet 4.6, narrate via ElevenLabs with word-level timestamps, render 9:16 clips with Kling 2.6 Pro on fal.ai, assemble and burn captions with FFmpeg, archive to Cloudflare R2, and upload through the YouTube Data API. A Next.js admin dashboard on Vercel manages topics, secrets, R2 storage, and pipeline runs.
Key highlights
- 8-step pipeline: topic → script (Claude) → narration (ElevenLabs) → video (Kling 2.6 Pro) → captions → upload
- Daily GitHub Actions cron with zero manual intervention
- Cloudflare R2 archival with deterministic naming
- Next.js admin dashboard for topic queue, secrets, R2 browser, and run history
- Telegram bot + Resend email notifications on completion or failure
- YouTube Data API v3 integration for upload, metadata, and thumbnail
Tech stack
- Next.js 16
- NeonDB (Postgres)
- Cloudflare R2
- GitHub Actions
- Claude Sonnet 4.6
- ElevenLabs
- Kling 2.6 Pro (fal.ai)
- FFmpeg
- YouTube Data API v3
- Resend
- Telegram Bot
- Vercel
Topics
- #ai-video
- #youtube-shorts
- #automation
- #anthropic-claude
- #elevenlabs
- #ffmpeg
- #content-pipeline
- #github-actions