4K 3840x2160 to 9:16: a 1214x2160 strip, then 1080x1920

Cropping a 3840x2160 clip to 9:16 keeps a strip 1214 px wide and 2160 px tall, which is larger than the 1080x1920 target in both directions, so scale=1080:1920 shrinks it by about 0.89x and the result stays sharp. The fractions are the same as for any 16:9 source: width 0.3164, x 0.3418, full height.

Why 4K is the good case

The crop is the one lossy step. With a 4K source it throws away pixels you do not need, while the final scale only reduces. That order, crop first and scale second, is what video-filter does: ops[] apply before filtergraph.

YouTube's own Shorts help page says a Short can be uploaded at a maximum resolution of 1080p (YouTube Help, read 2026-10-05). So there is no reason to keep a 2160 px tall file for Shorts; scaling to 1080x1920 is the right end state, and it also keeps the file small.

Order of work for a long 4K take

Video filter accepts sources up to 300 seconds. A 4K recording is often longer. Cut the range you want with video trim first (source up to 1800 s, output up to 900 s; precision: exact also accepts an output object to conform the size), then crop and scale that cut. Each step creates a new MP4 and leaves the source alone.

Decoding 4K is heavier than 1080p, and a program can pass the check and still fail on the box for memory or time; the docs say so explicitly, and a failed job returns a structured error. Keep the clip short where you can.

# 1) free contract check, no job, no charge
curl -X POST https://api.sume.com/v1/video-filter/check \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/uhd.mp4",
    "ops": [{"op":"crop","x":0.3418,"y":0.0,"width":0.3164,"height":1.0}],
    "filtergraph": "scale=1080:1920,setsar=1"
  }'

# 2) same body to /v1/video-filter, with an Idempotency-Key header, to encode

Checklist

At the documented public rate the encode is $0.02 per job and the check is free; confirm the live rate in GET /v1/catalog. The clip must already be a media.sume.com file of your workspace, so import it first with POST /v1/media-imports, and send an Idempotency-Key on the encode.

• Source 300 s or less for the filter step.
• Crop width 0.3164, x 0.3418, y 0, height 1.
• filtergraph is scale=1080:1920,setsar=1; stay under 2048 characters and 32 filters.
• Pull one still with video frames from the result to confirm the framing.

Pixel budget and job behavior

The strip holds about 2.62 million pixels and the target about 2.07 million, so the scale throws away roughly a fifth of what the crop kept. That is the healthy direction: the final picture is built from more source detail than it displays.

Plan for a heavy decode. 4K H.264 or HEVC takes more memory and time than 1080p, and the video-filter docs say a program that passes the check can still fail on the worker for memory or time, in which case you receive a structured job error. If you wait with mode: "sync", the handler holds the connection for at most 30 seconds, then returns 202 and you poll the job. For a long 4K file, trim first, then filter.

If you only have 3840x2160 footage that is 60 fps, the output keeps that frame rate unless the program changes it. Shorts do not need 60 fps for most content, and a lower frame rate makes a smaller file; you can set the rate in a later Timeline render with output.fps, but only if you accept the resample that comes with it.