A script for any video
Push in where the words lean
by Takehound
Pushes in on the lines that carry the talk, about once a minute, each held for the whole line. Chosen by what is said, never by a clock or at every cut.
For any video, 13 steps
Try it on my footageSign in with Google for a free first video
What Hound does with it
- Goal: push-ins a viewer feels and does not notice. Each lands on a line worth leaning into and holds for that whole line, and the words choose them. Push-ins at even gaps, or at every cut, are the plainest sign that a machine edited the video.
- Read their notes first. If they have taken push-ins off before or said they do not want them, place none, say so and stop.
- Read the whole transcript before choosing anything. List the lines with weight: the one a viewer would repeat, the reveal after a setup, a number that matters, the turn of the talk, a promise or a hard truth. Where two phrases belong together, the push goes on the half that carries the point, not the setup.
- Keep about one for each minute of talk, the strongest first, and one at most in a talk under a minute. Where two picks fall within 20 seconds of each other, keep the stronger. Then read the seconds you kept: if the gaps between them are nearly even, you chose by the clock, so choose again by the words.
- Each push-in starts on the first word of its line and ends in the breath after its last word. Never on one word alone: shorter than about 1.5 seconds it reads as a twitch. For a line longer than about 12 seconds, push in on its strongest sentence only.
- Read the timeline at each pick. Leave it out when a card, a clip or a picture of theirs covers the speaker there, when the frame shows a slide or a screen and not the speaker, or when that stretch is already pushed in (a scale above 1).
- Look yourself, with a still in the middle of each pick, at how the speaker sits in the frame. If pushing in would cut off the top of their head, their hands as they make the point, or anything they hold up, push in less or leave it out.
- Push in about 6 percent, scale 1.06, the same for every pick, so the framing never wanders. Only the single peak of the whole talk may take 1.1. Never more than 1.15.
- Lay them all with punch_in in one script, so one undo takes them all back. Leave every push-in already there as it is.
- If the talk already changes its framing at most cuts (a scale above 1 at every other join), do not take it off on your own: it may be hiding jumps that nothing else covers. Say how many times a minute it changes the framing, then stop and ask whether to keep it, keep it only where no card or clip covers the join, or take it off.
- When they answer, do only what they named: take a push-in off with punch_in at scale 1 over its seconds, one join at a time, and leave the ones you placed.
- Check: watch two seconds either side of each push-in. It begins as its line begins and lets go after the last word, no word is clipped or doubled where the clip was split, the speaker's head and hands stay in the frame, and no card sits half outside it. Fix only what you saw.
- Say how many push-ins you placed for how many minutes of talk, each with its second and the words it holds, the strong lines you left out and why, and that one undo takes them all back, or any one comes off by asking.
How it works
- Sign in with GoogleYour first video is free, a recording of up to 30 minutes.
- Open the editor on this scriptAdd your talk and your own footage. The editor opens on this template.
- Watch Hound workStop him at any point. Every change he makes can be undone.