· 5 min read
From a YouTube link to a new short: transcript, script, voice, video
A twenty-minute YouTube video usually contains three or four shorts. Getting them out is a chain of small jobs — transcribe, extract, rewrite, voice, shoot — and the chain is where the time goes, because each job's output is the next one's input and you will want to redo one of them.
Get a transcript you can trust
Auto-captions exist for most videos, but not all, and the ones that exist mangle names and numbers. Pull the captions when they are there; when they are not, run the audio through a speech-to-text model. Either way, read it once with the video playing — a transcript with a wrong number in it produces a script with a wrong number in it.
Extract the argument, not the summary
A summary tells you what the video was about. What you need is the claim it makes and the evidence it uses, because a short is one claim with one piece of evidence. Ask: what does this video want me to believe, and what is the single strongest thing it shows to make me believe it? That pair is your short.
Write the shorter script
Open with the claim, phrased as something the viewer did not expect. Give the evidence in the middle. Close with what to do about it. Forty to sixty seconds of speech is roughly a hundred to a hundred and forty words; write to that and cut anything that is context rather than payload.
- Claim first, in the first sentence
- One piece of evidence, shown rather than described where possible
- One instruction at the end
Voice it, then shoot to the voice
Record the voice-over first, or generate one, and only then plan the visuals. Timing the pictures to the words is much easier than the reverse. For each sentence decide what is on screen — a screenshot of the source, a generated clip, text — and hold it for as long as the sentence lasts.
Keep the steps connected
The reason this takes an afternoon by hand is not any single step; it is that the transcript lives in one tool, the script in another, the voice in a third and the cut in a fourth, and a change in the script means walking the whole chain again. On a canvas the chain is drawn: the YouTube link is a source node with the transcript inside it, the script is a node fed by it, the voice by the script, the video by both. Edit the script and re-run only the voice. That is what Infica's AI video generator is built to do; the script generator page covers the writing step on its own.