How AI clip generation actually works

Every Ready-to-Post clip in a Content Kit went through the same three-step pipeline: transcribe, select, render. Here's what actually happens at each step, and why it's built the way it is.

1. Transcribe

The source video's audio is transcribed with word-level timestamps — not just what was said, but exactly when each word starts and ends. That precision is what makes accurate, tightly-timed captions possible later, instead of captions that drift out of sync with speech.

2. Select

The full transcript is analyzed to find the strongest short-form moments — real speech boundaries, not arbitrary time slices. Each candidate gets a Viral Potential score out of 100, broken into Hook, Flow, Value, and Trend, with a one-sentence rationale for each. This is an estimate of short-form performance potential, not a promise — it's a starting point for a Clipper's own judgment, not a replacement for it.

3. Render

Each selected moment is trimmed, cropped to a vertical aspect ratio, and rendered as its own clip with captions burned in — short bursts of a few words at a time, in the style you choose, not one long block of text sitting on screen. The result lands in the Content Kit's Ready to Post tab, downloadable as-is or editable first.

Why it's built this way

The goal isn't to replace editorial judgment — it's to remove the blank-page problem. Starting from five AI-scored candidates with real captions already in place is a faster way into a campaign than starting from a raw, uncut video file.