Fake a second camera on one AI clip: a punch-in cut for Shorts

How do you cut between two angles when you only have one clip?

Crop a copy of the clip to a tighter frame, then alternate between the original and the copy in a Timeline 1.0 render, using the same moments of the clip for both. The viewer sees a wide shot, a closer shot and a wide shot again, with no regeneration. On Sume that takes three pieces: a crop from video filter, a spine of audio from audio detach if the clip carries its own sound, and a render.

The effect is a punch-in, not a real second camera. The tight shot is the same pixels enlarged, so it suits a talking head or a product on a desk and does not suit fine detail.

Crop by fractions, then check it for free

The crop op takes a rectangle as fractions of the source frame: x and y between 0 and 1, width and height between 0.05 and 1, with x + width and y + height no more than 1. A centred 80 percent crop is x: 0.1, y: 0.1, width: 0.8, height: 0.8. For a face in the upper half of a vertical frame, pull y up, for example 0.05, so the head is not cut off.

POST /v1/video-filter/check runs the same validation as the encode and returns diagnostics without creating a job or reserving credits. Run it first. The encode is a flat $0.02 per job in the docs; the check is free. The crop keeps the duration and the audio of the source, so the cropped file lines up with the original moment by moment.

curl -X POST https://api.sume.com/v1/video-filter \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: punch-in-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/clip.mp4",
    "ops": [{ "op": "crop", "x": 0.1, "y": 0.05, "width": 0.8, "height": 0.8 }]
  }'

What the crop costs in sharpness

Timeline 1.0 fits each slot to the output frame, and the default fit is cover. A tighter crop therefore gets enlarged more. The numbers below are plain arithmetic for a 720 by 1280 clip going into a 1080 by 1920 Short, with the docs noting that the compiler rounds crops to even sizes.

Alternate the shots in the render

In the render, each video[] slot has an on-spine start and a source_in into its own file. Give the tight slot the same source_in as its start so it shows the moment the voice is on. The slots below cut hard at 4 and 8 seconds; no transition means no fade, and a hard cut is what makes a punch-in read as an edit. The audio spine is the detached track of the original, so the voice never jumps.

Pricing in the docs is $0.10 per ceil(output minute), so this 12-second Short is one billed minute. With a $0.01 detach, a $0.02 crop and the render, the fixed-rate steps come to $0.13. That figure excludes generating the original clip.

{
  "audio": { "url": "https://media.sume.com/artifacts/artf_demo/voice.wav", "duration_seconds": 12 },
  "video": [
    { "source_url": "https://media.sume.com/artifacts/artf_demo/clip.mp4", "start": 0, "duration": 4 },
    { "source_url": "https://media.sume.com/artifacts/artf_demo/clip-tight.mp4", "start": 4, "source_in": 4, "duration": 4 },
    { "source_url": "https://media.sume.com/artifacts/artf_demo/clip.mp4", "start": 8, "source_in": 8, "duration": 4 }
  ],
  "output": { "width": 1080, "height": 1920 }
}

Where this falls short

A punch-in only changes framing. It will not change the lighting or the angle, and a clip with a moving camera may drift out of the cropped frame. Leave output.fps unset so the render matches the source rate; a rate that differs makes the compiler repeat or drop frames and warn with output_fps_resamples_sources, as the frame rate post explains.

YouTube's help for the Shorts camera describes recording in segments with a progress bar. That is the in-app version of this idea, where you cut between takes. Here you build the cut after the fact from one take. Check the filter program before paying with the free check endpoint post.