Street interview, rebuilt for a new subject

A real reference video, the structure that transfers, and the command that rebuilds it with your own content.

The reference

The reference is a vox-pop interview: a repeated question, four strangers, escalating answers, subtitles burned in. What transfers is the rhythm — question, answer, cut, repeat — not the people.

Structure, as cloned

  1. 1. Cold open

    Starts mid-answer; the question only appears as a caption.

  2. 2. Four answers

    6–10 seconds each, cut on the last word.

  3. 3. Escalation

    Answers get stranger; the loudest is held for last.

  4. 4. Text payoff

    Question restated as a caption so it reads on mute.

What changed

  • — Presenter and voice
  • — The question asked
  • — Every answer
  • — B-roll of the location
  • — Caption wording

What the format keeps

  • — Question–answer cadence
  • — Answer order (mild → surprising)
  • — Caption position and size
  • — Cold open on an answer, not a setup

Run it yourself

npx skills add hypit-ai/hypit -g
hypit run clone --reference ./reference/interview.mp4 --script ./interview.svs --run

The reference file stays fixed; only the script changes between runs. That is what makes the format reusable.

Keep reading

FAQ

Can I run this format without on-camera strangers?

Yes. One presenter can play both sides, or the answers can be generated from a script — the retention comes from the question–answer rhythm.

Why does the reference matter so much here?

Because the pacing is the product. The same script delivered with slower cuts loses the format entirely, which is why we clone the reference instead of describing it.

Where do the captions come from?

They render from the caption track against the script, so re-recording an answer does not desynchronise them.