Can video-trim output a 4K frame? Each side is capped at 2160 pixels

No. The output object on video trim takes width and height between 256 and 2160, so a 3840-pixel side is outside the range. You can conform a clip to something like 1080 by 1920 or 2160 by 1215, at 24, 25, 30 or 60 fps, but not to a 4K frame.

This matters this week because new video releases advertise 4K. Vidu's Q4 Preview announcement on October 7 lists output from 540p to 4K (PR Newswire release, read 2026-10-10). A 4K master is a fine thing to own; Sume's trim conform is simply a down-conform tool.

What the output object accepts

The conform is part of the same re-encode as the cut, so you pay one $0.02 job and not two. It happens once, with libx264 and yuv420p, and the kept audio is remuxed as AAC.

Why keyframe cannot conform

precision: "keyframe" is a stream copy. It starts at the nearest earlier keyframe, so the cut can begin a GOP early, and the result reports actual_start_seconds so you can re-base your times. A stream copy cannot change the frame size or the frame rate, which is why output is exact only.

If you need both a quick cut and a conform, use exact. If you only need a fast cut and the geometry is already right, use keyframe and leave output out.

The inputs you still need

Trim takes one Sume-hosted clip: video_url must be a media.sume.com artifact or asset in your workspace, start is at least 0, and you send exactly one of end or duration. The source can be up to 1800 seconds and the output up to 900 seconds, with a minimum of 0.2 seconds. An end past the source clamps and returns the warning trim_clamped_to_source.

A clip from another vendor is not accepted by URL on this route. Off-host addresses are rejected at admit with unsupported_media_source, and the clip has to be on the Sume media host first.

Choosing a conform target

For a vertical feed, 1080 by 1920 sits well inside the cap. For a landscape 4K deliverable, no trim output setting reaches it, so conform to a 2160-pixel side only when that is the real destination size. Remember that conform shrinks or reshapes the frame; it adds no detail. If a draft is small and you need more pixels, the separate video upscale route is built for that and is priced per input second.

Frame rate deserves the same care. If two clips in a Timeline render have different native rates, the render repeats or drops frames and warns output_fps_resamples_sources. Conforming each clip to one of the four allowed rates in trim, then rendering at that rate, avoids that judder.

Cost check

A trim is $0.02 whether you conform or not. Ten clips cost 10 x $0.02 = $0.20, and 100 cost $2.00. Confirm the live rate in GET /v1/catalog and send an Idempotency-Key on each write so a retry cannot create a second job.

Refusals are explicit. video_trim_range_required means neither end nor duration was sent, video_trim_range_conflict means both were, and video_trim_range_empty means end is not after start or the range is longer than 900 seconds. A request that carries ffmpeg-style fields such as vf, codec or crf returns ffmpeg_fields_rejected, because the server compiles the ffmpeg command itself and the output object is the only way to shape the result.