Crop 9:16 to 2:3 for Pinterest: height 0.84375, then 1000x1500

To turn a 1080x1920 (9:16) clip into 2:3, keep the full width and crop the height to 0.84375, which is 1620 of 1920 pixels. Centered, y is 0.078125. Pinterest's ad specs page lists 2:3 for Showcase and Collections, with 1000 x 1500 recommended, so the last step is a Timeline render at that size.

The numbers

A 2:3 frame at 1080 wide is 1620 tall. 1620 divided by 1920 is 0.84375. The most you can shift the window is 0.15625, which is 300 pixels, so the crop removes 300 pixels in total from the top and bottom.

The request

Crop fractions are applied to the source frame, so the pixel result depends on the source size. The video filter docs require x and y in [0, 1], width and height in [0.05, 1], x + width at most 1 and y + height at most 1, and the compiler rounds to even values. One crop is one $0.02 encode job, the check is free, the source must be 300 seconds or shorter, and the request needs an Idempotency-Key. Sume does not auto-detect faces or subjects, so you choose the offset.

curl -X POST https://api.sume.com/v1/video-filter \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: crop-2x3-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
    "ops": [{ "op": "crop", "x": 0, "y": 0.0781, "width": 1, "height": 0.8437 }]
  }'

Scale to the recommended size

1080x1620 is 2:3 exactly. Pinterest recommends 1000 x 1500, which is also 2:3, so Timeline 1.0 can scale down with output.width 1000 and output.height 1500 and nothing is cropped further. Scaling down is lossless in framing, which makes this the cleaner direction than going up.

What Sume does not do

Sume does not pick the crop window for you and does not publish to Pinterest. Standard video on the Pinterest page also accepts ratios between 1:2 and 1.91:1, so a 9:16 clip is allowed as is; crop to 2:3 only when you want the Showcase or Collections shape.

Chain the steps in order: crop with video filter, put the returned MP4 into a Timeline video slot, set the 1000x1500 output, then add captions last if you want them. The filter job costs $0.02 and the render $0.10 per output minute, rounded up, so a 15-second pin costs about $0.12 in Sume jobs, not counting any model generation that made the clip.