Trim before you recast: pay for the seconds the swap needs

Recast bills the full length of the source video, rounded up to the second, so the cheapest swap is the shortest source that still contains the shot you need. Cutting a 30-second clip to the 12 seconds where the person appears takes the 768p price from $11.25 to $4.50 on Sume ($0.30 list per second times 1.25, read 2026-10-03). The trim itself is a separate, small job: POST /v1/video-trim returns a new MP4 and leaves the source untouched (Video trim docs).

Two limits make this more than a saving. The source must be 5 to 30 seconds, and no single shot may exceed 15 seconds, so trimming is also the way to get a long raw take inside the envelope.

The arithmetic

Because the price is linear in seconds, savings are linear in the part you drop. The table uses fal's list prices times Sume's 1.25 and rounds up to the cent; the trim job is a Modal-compute charge that is small next to a video render, and the docs describe it in the trim page.

Trim, then recast

Trim takes a video_url that is a media.sume.com artifact or asset in your workspace, a start in seconds and exactly one of end or duration. There is no open-internet fetch, so a clip hosted elsewhere is imported first with POST /v1/media-imports. Idempotency-Key is required. The result is a new MP4 URL that you pass straight into Recast.

# cut 12 seconds starting at 6.0 s
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-001" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_demo/raw.mp4","start":6,"duration":12}'

# then recast the trimmed MP4 (URL from the trim result)
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: recast-trim-001" \
  -d '{"model":"h3-max-recast","video_url":"https://media.sume.com/artifacts/artf_demo/trimmed.mp4","reference_image_urls":["https://example.com/new-host.jpg"],"resolution":"768p"}'

Where to cut

Choose the range with the cuts in mind. Recast keeps the source's cuts, so a trim that lands in the middle of a shot gives you a clip that starts and ends mid-action. That is fine for a social cut and awkward for a hard edit. To find the edges, sample stills from the clip with video inspect, which can return stills at timestamps you choose and probe facts such as duration and whether the clip has audio. Look at a handful of stills around the candidate range and pick a start and end where the person is fully in frame.

The 5-second floor still applies after you trim. A 3-second range is a perfectly good trim but is below the Recast minimum, so extend it to at least 5 seconds. And the 15-second shot rule still applies: if the person stands in one unbroken shot for 20 seconds, trimming to 14 seconds of it satisfies the rule, while keeping all 20 would not.

A worked example

Take a 30-second raw interview where the presenter you want to replace is on screen from 6 to 18 seconds and the rest is a wide shot of an empty room. Recasting the whole clip at 768p costs $11.25. Trimming to 6 through 18 gives a 12-second source; the Recast is $4.50, a saving of $6.75, and that holds for every variant you produce from the same trim. If you make six market variants, the trim pays for itself six times: $67.50 untrimmed against $27.00 trimmed at 768p.

The arithmetic gets better at 1080p. Six variants of the 30-second clip are $101.28 at 1080p, against $40.50 for the 12-second trim. Draft at 768p, finish at 1080p, and trim before both.

When trimming costs more than it saves

Trim is not free, and it adds a step and a failure point. If the whole 30-second source is the deliverable, skip it. If the source is already 8 seconds, the saving is nothing. The sweet spot is a long raw capture where the person occupies a fraction of the time, which is common for interview b-roll, product demos and UGC raw files.

A last note on audio. By default the audio-handling parameters on trim are optional, so check the trim page for the audio field before you rely on a particular behavior, and re-listen to the trimmed clip: Recast keeps whatever audio the source carries.