Face swap beta takes no aspect ratio: crop 16:9 to 9:16 first ($0.02)

Sume's Avatar Face Swap beta has no aspect-ratio field, so the output keeps the shape of your source. To get a vertical result from a 16:9 clip, crop the source to 9:16 first with the video filter route ($0.02 per job), then run the swap. At the plus tier that is $0.02 + $3.675 (the 15-second reservation) = $3.695.

The crop fractions

The video filter crop op takes x, y, width and height as fractions of the source frame. A 9:16 slice of a 16:9 frame is the full height and (9/16) / (16/9) = 0.3164 of the width. Centering it puts x at (1 - 0.3164) / 2 = 0.3418. On a 1920 x 1080 source that is a crop of about 608 x 1080, so expect lower pixel density than a native vertical shoot.

Vertical is the usual reason, but the same trick works for 1:1. A square slice of a 16:9 frame is 0.5625 of the width at full height (9/16), centered at x = 0.21875. Whatever the target shape, the crop never adds pixels, so the output resolution is the source height times the new width.

{
  "video_url": "https://media.sume.com/artifacts/artf_demo/wide.mp4",
  "ops": [
    { "op": "crop", "x": 0.3418, "y": 0, "width": 0.3164, "height": 1 }
  ]
}

Order of operations

The video filter route works on one clip already on media.sume.com, so import the source first. The docs require an Idempotency-Key on the encode, and the free check endpoint validates the program before any charge. The result is a new MP4 with duration, and its URL is what you pass to the swap as video_url, subject to the swap's rule that the URL must be a fetchable public HTTPS video.

Plan for the extra step in a batch. For 50 swapped variants at plus, the crop adds 50 x $0.02 = $1.00 on top of 50 x $3.675 = $183.75, a 0.5 percent overhead. That is small enough to apply to every source rather than checking each one by hand.

Keep the face in the slice

A centered crop assumes the face is centered. If the subject sits left or right in the wide frame, move x so the face stays inside the slice. Run the free check, encode once, look at the result, and only then spend on the swap. The source clip must still be 4 to 15 seconds with usable audio, and the filter keeps the source audio.

After the crop, check the result before you commit to the swap. Open the cropped MP4, confirm the face is inside the frame for the full clip, and confirm the duration is still within 4 to 15 seconds. The filter keeps the source audio and frame rate unless the program changes them, so the swap still sees usable audio. If the subject moves across the wide frame, a fixed crop will cut them off; pick a source where they stay in place, or choose a different shot.