04 / Automation · AI — 2026
AI Video Content Pipeline
Automated short-video production system — scripts, footage, karaoke captions and scheduling across platforms.
- Role
- Architect + operator
- Stack
- Python, LLM APIs, Pexels API, FFmpeg
- Year
- 2026
- Status
- Shipped & live
(01) The problem
Producing daily short-form video for multiple channels (football, quotes) is a full-time editing job: sourcing footage, writing scripts, captioning, timing music, exporting per-platform formats. Manually, it doesn't scale past one channel.
(02) What I did
- 01
Designed a pipeline with style presets per channel: LLM-generated scripts, licensed footage pulled from the Pexels API, karaoke-style captions and music-level control.
- 02
Built auto-cleanup and scheduling so the system runs as a production line, not a demo.
- 03
Evaluated third-party avatar generation (D-ID), measured quality against cost, and rejected it — knowing when not to buy is part of coordination.
(03) The outcome
Two channel formats in production; five videos scheduled from a single run.
30+ test videos produced during development across five channel concepts.
One person now operates what is effectively a small content studio.
- 5
- Videos per run
- ≈0 h
- Editing time
- 2
- Channel formats