Hour-long podcast video to clips: the 1800-second source cap
Sume's trim, detach and inspect tools read sources up to 1800 seconds, so a 60-minute episode must be split before import. Limits and a safe split plan.

You cannot feed a 60-minute podcast video to Sume's trim, detach or inspect tools in one piece: each reads a source of at most 1800 seconds (30 minutes) and fails longer ones as source_duration_exceeded. Split the episode into halves in your editor, import both, and cut clips from each half.
Which limits apply to a podcast clip pipeline?
A 60-minute episode is 3600 seconds, double the cap on the first three tools. Video filter and compose are tighter still at 300 seconds, so clip first, then filter.
| Tool | Source cap | Output cap | Rate |
|---|---|---|---|
| Video trim | 1800 s | 900 s, at least 0.2 s | $0.02 per job |
| Audio detach | 1800 s | 900 s (use range) | $0.01 per job |
| Video inspect | 1800 s | 24 stills per call | Probe and stills free; transcript $0.01 per audio minute |
| Video filter | 300 s | inherits source | $0.02 per encode |
| Video captions | public HTTPS URL | fixed estimate for up to 60 s | $0.20 per job |
How should you split the episode?
Sume has no tool that splits a source longer than its own cap, so do the split outside it, in the editor you already use. Cut at a natural break, such as an ad read or a topic change, and keep both halves under 1800 seconds. Then overlap: let the second half start about 30 seconds before the first one ends, so a clip that straddles the break survives in at least one half.
Import each half with POST /v1/media-imports, since the tools read this workspace's media.sume.com media, not the open internet.
How do you find the moments?
Run Video inspect on each half with transcribe: true and sentence segmentation. The transcript gives you timed lines to scan for a hook or a quotable answer. The duration_seconds hint is capped at 600 and omitting it reserves one minute of transcription.
How do you cut and caption each clip?
Cut with Video trim using start plus exactly one of end or duration. Exact precision is the default and is frame-accurate; keyframe is a stream copy that can start a GOP early, so re-base against actual_start_seconds.
Then caption the short clip, not the half. The standalone caption price is fixed at $0.20 for videos up to 60 seconds, so clips of 45 seconds fit under it.
curl -X POST https://api.sume.com/v1/video-trim \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: ep42-half1-clip3" \
-d '{
"video_url": "https://media.sume.com/artifacts/artf_demo/ep42-part1.mp4",
"start": 612.4,
"duration": 45
}'What does a ten-clip episode cost?
At the published rates, ten clips cost 10 trims at $0.02 plus 10 caption jobs at $0.20, which is $2.20, before the transcript charge on the two halves. Rates can change; the catalog is the source of truth.
Two things to watch: the trim result's video_url is a new artifact each time, and you pass that URL on to the caption job, never the source half. And a source with no audio fails detach as detach_source_has_no_audio, so probe first if a half came from a screen recording.
How do you keep the batch tidy?
Name idempotency keys after the episode, the half and the clip, such as ep42-half1-clip3, so a retry returns the same job instead of billing a second cut. Keep a small table of half, start, duration and the resulting trim video_url; the clip's start inside the half is not the start inside the full episode, so add the half's offset when you want to cite a timestamp.
If a clip needs captions in the show's house style, caption it after the trim and reuse the same style and design for every clip, so a week of clips looks like one series.
What does Sume not do here?
It does not pick the best moments for you in this flow, it does not reframe a wide shot to a vertical speaker crop in these tools, and it does not split an over-long source. Those parts stay in your editor or your own script.
Sources
Related posts
More in Use cases
- How many clips for a 10-minute faceless video? Timeline slot math
A 10-minute faceless video fits Timeline 1.0 comfortably: up to 200 slots, a render near $1.00, and limits on fades and single-pass renders to plan around.
- Instagram Reels AI translation: Korean and Japanese added July 2026
Meta's July 14, 2026 update adds French, German, Italian, Japanese and Korean to Reels translation on Instagram. Here is the full language list and the dates.
- Korean captions on product clips: use korean-ad, not slam
slam, punch and tiktok-green have no Hangul glyphs and return 400 on Korean copy. Use korean-ad with language ko for Korean product clips; $0.20 a job.
- Find product moments in a live-commerce replay and trim them out
Run one transcript on a live-selling replay, find the sentences where each product is pitched, then cut each range with video-trim at $0.02 a clip.
Written by Sume