Flow video edit limits: 60 s, 1 GB, 10 s window, and the Sume route

Flow accepts an uploaded video up to 60 seconds long and 1 GB, asks you to trim anything longer than 30 seconds, and then lets you edit a segment of up to 10 seconds. If you want the same result through an API, the Sume equivalent is one video_url edit job on gemini-omni-flash-1.1, with the clip cut to 10 seconds first.

Those three numbers (60, 30 and 10) are easy to mix up, so this page lays them out as they appear on Google's help page and then maps each step to a Sume call.

What are Flow's upload and edit limits?

Google's Flow help page for editing and building scenes says you can upload, drag or select a video up to 60 seconds long and 1 GB. It adds that a video longer than 30 seconds must be trimmed, and that in the trimming window you select up to a 10-second segment. It also says edit and refine need Gemini Omni Flash and that the feature is only available in certain countries.

The Gemini API page for Omni Flash uses the same 10-second figure for the API: input videos for editing or extending are 10 seconds at most. So whether you work in Flow or call the API, the unit of work is a clip of ten seconds or less.

How do I prepare the same clip for Sume?

Sume's file-handling tools read clips that already live on media.sume.com, so import the file first with POST /v1/media-imports, then cut it. Video trim takes a video_url, a start, and exactly one of end or duration, and returns a new MP4 without touching the source.

Video inspect returns the probe facts for a clip, including whether it has audio, and its docs cap the source at 1,800 seconds, which is far past anything you would send to an edit. Use a probe-only call (frames: false) when all you need is the duration.

curl -X POST https://api.sume.com/v1/video-trim \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: trim-for-omni-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
    "start": 12,
    "duration": 8
  }'

What does the edit call look like?

Take the trimmed artifact URL from the trim job's result and send it as the edit source. The Video Router page documents video_url as the edit source for gemini-omni-flash-1.1: the prompt describes the change, resolution is optional with a 720p default, and aspect_ratio and duration are not sent in edit mode. video_url cannot be combined with image_url, end_image_url or the reference fields.

Native audio is always on for this model, and generate_audio: false is rejected, so the edited clip comes back with sound.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: omni-edit-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Remove the person on the left. Keep everything else the same.",
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk-trimmed.mp4",
    "resolution": "720p",
    "mode": "async"
  }'

What does Sume not do that Flow does?

Sume has no visual trimming window, no Insert, Remove or Camera mode switch, and no History panel. An edit is a prompt plus a source URL, and each job is a new request. Flow's page does not state a price for an edit, and I am not comparing money here; for what a Sume Omni job bills, the Video Router page says provider list times 1.25 per output second by resolution.

One practical difference: Flow's 30-second trim prompt is a hard step in the interface. With the API nothing forces you to trim, so a 40-second source sent straight to an edit is your mistake to catch. Probe first, cut second, edit third.

Sume's docs do not state an input-length cap for the edit source on this model, so follow Google's 10-second limit rather than testing the edge. For the full job lifecycle (submit, poll, result) see Jobs and results.

Which route should I pick?

Use Flow when one person is iterating by eye on a handful of clips and wants the Scenebuilder and History panel. Use the API route when the same edit runs over many clips, when a script owns the trim, or when the result has to land in a pipeline as a durable URL.

Either way, trim to ten seconds or fewer before you ask Omni for anything.