Skip to content
All work
In-house productWeb App · Media Automation2026

Viral Clipper

Turn one 8-hour stream into a week of ready-to-post vertical clips — automatically, for the price of a few API tokens.

Built with

  • Python
  • yt-dlp
  • faster-whisper
  • Gemini
  • ffmpeg
  • Flask

Services

  • Product design
  • Full-stack build
  • ML/media pipeline
  • Workflow automation

The outcome

8h → clips
From full VOD to shorts
9:16
TikTok / Shorts-ready
Token-only
No per-clip SaaS fees

Overview

Viral Clipper is our self-hosted answer to the £100+/month "clip-it-for-you" tools. Feed it a long-form stream and it returns finished, captioned, vertical clips ready to post — running on your own machine, so the only thing you pay for is the AI tokens it actually uses. We use it ourselves every week — it turns an afternoon of manual clipping into a hands-off batch.

The challenge

Clippers face the same grind: scrub hours of footage, hunt for the moments that land, reframe everything to vertical, caption it, export. The tools that automate this rent the workflow back to you at a steep monthly price and still keep your footage on their servers. We wanted the same output without the subscription or the lock-in.

The approach

We built an end-to-end pipeline that runs locally and only reaches out to a paid API for the one job worth paying for — judgment about what’s actually interesting. Everything else runs on open tooling on your own hardware.

The pipeline starts with yt-dlp, pulling the source VOD straight from the platform. Audio is transcribed locally with faster-whisper on the GPU, giving us a timed transcript to reason over without sending anything to a third party.

That transcript goes to Gemini, which ranks the stream for the moments most likely to clip well — the spikes, the payoffs, the quotable lines — and returns candidate in/out points. We layer a feature-based detector on top to keep the subject framed: it tracks the on-screen presenter and drives the vertical crop so faces never drift out of frame.

ffmpeg does the cutting, reframing to 9:16, and burns in captions. The whole run is driven from a Flask dashboard where you drop in a VOD, watch the pipeline work, and pull down the finished clips.

The result: an 8-hour stream becomes a batch of post-ready shorts with no manual scrubbing, and the running cost is a handful of API tokens instead of a monthly bill.

The dashboard — VOD in, clips out
Moment ranking from the transcript
Auto-reframed 9:16 output with captions
Start a project

Want results like In daily use — saves hours of editing every week?

Tell us about your business and what you need. We reply to every enquiry within two business days.