Hide jump cuts with alternating punch-ins: one crop copy, N slots

Pause cuts leave the speaker in exactly the same pose on both sides of a join, which reads as a jump. The usual short-form fix is to alternate between the full frame and a slight punch-in; with Sume you make the punched copy once with a $0.02 video-filter crop (width and height 0.8 of the frame is a 1.25x zoom), then give each Timeline slot either the wide or the punched file with the same source_in.

Instagram's First Draft cuts the pauses for you in one tap (read 2026-10-09); this is what you add on top when you script the cut yourself.

The crop

Crop fractions are relative to the source frame: x and y in [0, 1], width and height in [0.05, 1], with x + width and y + height at most 1. A centered 0.8 x 0.8 window starts at 0.1. It keeps 0.8 x 0.8 = 0.64 of the pixels, discarding 36%, so check it with the free POST /v1/video-filter/check before you encode. The punched file has a smaller frame than the wide one; Timeline slots default to fit: "cover", which fills the output canvas, so it simply looks closer. Confirm with one still.

{
  "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
  "ops": [{ "op": "crop", "x": 0.1, "y": 0.1, "width": 0.8, "height": 0.8 }]
}

Cost against filtering every slot

The punched copy covers the whole clip, so Timeline source_in simply picks the part you need. Assumption: a 14-slot cut, seven of them punched.

Gotchas

• Both files must stay in sync, so give a wide slot and a punched slot the same source_in, and use one audio spine for the whole Reel.
• Cropping 36% of the pixels softens the picture on a 1080p source; check one still before you ship.
• The filter source must be 300 s or shorter, and the call needs an Idempotency-Key.
• Keep punch-ins to a mild 1.25x; the filter cannot scale beyond what the crop window keeps.