Overview: This video demonstrates building an end-to-end, local AI-powered clipping pipeline using Claude Code as orchestration. It converts long-form videos into vertical short clips by performing face detection, tracking, speaker detection, framing, captioning, and rendering locally with open-source tools—eliminating per-clip API costs. The presenter walks through a three-stage build plan, tests on multi-speaker footage, and discusses limitations and iterative fixes.
- Stage breakdown: The pipeline is split into face detection/tracking, speaker detection and crop planning, and final rendering with captions—each built and tested separately for easier debugging.
- Orchestration, not processing: Claude Code acts as the “brain” that sequences open-source tools and prompts; actual video processing runs locally on your machine.
- Local + open-source: No paid clipping API or per-clip costs; includes a debug mode that overlays tracking boxes to verify framing and speaker decisions.
- Practical use and limits: Suited for podcasts/interviews and scalable clipping workflows; may require follow-up prompts or manual edits when speakers overlap, move rapidly, or camera angles are unusual.
Quotes:
Why don’t we just build the whole thing ourselves?
You give it just one long video, and it gives you a finished vertical clip.
It’s not just a clipping tool anymore. It can actually become a scalable service.
Statistics
| Upload date: | 2026-08-25 |
|---|---|
| Likes: | 79 |
| Comments: | 13 |
| Statistics updated: | 2026-08-27 |
Specification: I Rebuilt a $99/Month AI Clipper With Claude Code
|
