Crop 16:9 to 9:16, 1:1, 4:5, 3:4 and 2:3: video-filter fractions
Video filter crop takes fractions of the frame. For a 16:9 source, a full-height crop is width 0.3164 for 9:16, 0.5625 for 1:1 and 0.45 for 4:5. $0.02 each.

To crop a 16:9 video to another aspect ratio with Sume's video filter, send a crop op with y: 0, height: 1, and a width of the target ratio times 9/16, with x set to half of what is left over. For 9:16 that is width 0.3164 and x 0.3418; for 1:1 it is 0.5625 and 0.21875; for 4:5 it is 0.45 and 0.275.
The op takes fractions of the source frame, not pixels, so the same numbers work for 1280x720 and 3840x2160 sources. Each encode is $0.02 and the program check is free. The rules below come from the Sume video filter page, read on 2026-10-03; the arithmetic is mine.
The fraction table
The docs define crop as a rectangle in fractions of the source frame: x and y in 0 to 1, width and height in 0.05 to 1, with x + width <= 1 and y + height <= 1. For a full-height crop of a 16:9 frame, the width fraction is the target aspect ratio divided by 16/9, which is the same as multiplying by 0.5625. Centering puts x at (1 - width) / 2.
The compiler rounds the result to even pixel sizes for yuv420p, so a 1920x1080 source cropped at 0.3164 lands within a pixel of true 9:16. If you need an exact output size, set it on the timeline render that follows, where output.width and output.height are explicit.
| Target | width | x (centered) | Pixels from 1920x1080 (before even rounding) |
|---|---|---|---|
| 9:16 | 0.3164 | 0.3418 | about 607 x 1080 |
| 2:3 | 0.375 | 0.3125 | 720 x 1080 |
| 3:4 | 0.4219 | 0.2891 | about 810 x 1080 |
| 4:5 | 0.45 | 0.275 | 864 x 1080 |
| 1:1 | 0.5625 | 0.21875 | 1080 x 1080 |
One request
Check first. POST /v1/video-filter/check runs the same schema and source preflight as the encode and returns diagnostics instead of a 400, without creating a job or reserving credit. A program that passes can still fail on the worker for resource reasons, which comes back as a structured job error.
The encode needs an Idempotency-Key and a media.sume.com source in your workspace, so import an outside file first. The default mode is async; mode: "sync" waits up to 30 seconds.
curl -X POST https://api.sume.com/v1/video-filter \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: interview-9x16-001" \
-d '{
"video_url": "https://media.sume.com/artifacts/artf_demo/interview.mp4",
"ops": [{ "op": "crop", "x": 0.3418, "y": 0, "width": 0.3164, "height": 1 }]
}'When the speaker is not centered
A centered crop is wrong for a two-person interview or a speaker standing on the left third. Move x toward the subject but keep x + width at or below 1; for a 9:16 crop that means x can range from 0 to 0.6836. Pull a still first to find the subject: the video frames route returns an exact frame at a time, and inspect returns up to 24 stills.
Five ratios from one interview is five encodes, or $0.10 in total. If a single clip has to feed all five placements, that is cheaper than re-shooting, and cheaper than generating new video. If the clip is already vertical and only needs a safe margin, a timeline fit of blur or cover handles it without a crop job.
- Width fraction for a full-height crop = target ratio x 0.5625.
- Centered x = (1 - width) / 2.
- Maximum x = 1 - width.
- Minimum width and height fraction: 0.05.
Sources
Related posts
More in Media tools
- Add captions to a recorded AI avatar call: what Sume needs
Tavus can record a live avatar call to your bucket. Sume's video captions need a public HTTPS URL and priced for clips up to 60 seconds. Steps and limits.
- Add subtitles to a video in DaVinci Resolve, or with a caption API
Resolve 21 transcribes and translates speech to text in the app. Sume video-captions burns styled captions into a public MP4 URL for $0.20 up to 60 seconds.
- AI music from an image: Lyria 3.5 image_url mood still on Sume
Sume's Music Router takes an optional public HTTPS image_url as a mood still next to the prompt. How to send it, what it costs and what to expect back.
- AI music generator for covers: what Sume cannot do, and a swap
Sume cannot cover an existing song: no audio input and no song-to-song path. Licensed cover platforms are still in development. Here is an original-track swap.
Written by Sume