I made a Telegram bot that gives me a fully rendered video, and it only cost me $0.032 to make
TL;DR: I send two terms to a Telegram bot ("Suspension bridge vs Arch Bridge"). A few minutes later it sends back a finished vertical video with a stickman explaining the difference, a voiceover, word-by-word captions and photos. Each video costs ~$0.032 (Claude API + Eleven Labs API). Yes, that's the only cost.
(Sorry for the music at the start - i felt it was too quiet so i added what was available)
What my automation does:
- I message my Telegram bot two things, e.g. Suspension bridge vs Arch Bridge
- The AI writes a short script and sends it to me
- I reply
APPROVE(orREVISE: make the hook punchier) - It records the voiceover, finds photos and builds the video
- The finished video comes back in the same chat
Stack
- n8n, self-hosted in Docker (
n8n_datavolume for workflows, credentials and data tables) - cloudflared quick tunnel to expose n8n over public HTTPS for Telegram webhooks
- Claude Sonnet 4.6 (Anthropic API) for script generation and revisions
- ElevenLabs
eleven_multilingual_v2via the/with-timestampsendpoint - Supabase Storage: public buckets for audio and rendered video
- Pexels API: one portrait image per term
- Remotion render server: Express + TypeScript, running on the host
Happy to answer all your questions!!