Weekly video series on Sume: Format run, captions, trim, timeline cost
A weekly series chains a Format run into captions, a trim and a timeline. Three finishing steps have fixed prices, so only generation needs a cap. The sums.

A weekly series is one Format run plus three finishing steps with published flat prices: a trim at $0.02, standalone captions at $0.20 for a clip up to 60 seconds, and a timeline render at $0.10 per started minute. Only the generation inside the Format run varies, so that is the only step that needs a spend cap.
The chain
Each step takes the previous step's media.sume.com URL. The Format run returns primary_output_url and artifacts[]. Trim cuts a range out of one clip and returns a new MP4. Captions burn text onto a public video URL. Timeline joins clips on an audio spine and returns one MP4.
None of these three finishing steps calls a generation model. Trim and timeline run ffmpeg on the worker media runtime, so their price is fixed in advance and does not move with the clip.
| Step | Endpoint | Public rate |
|---|---|---|
| Cut a range | POST /v1/video-trim | $0.02 per job |
| Burn captions (up to 60 s) | POST /v1/video-captions | $0.20 per accepted job |
| Assemble clips on a spine | POST /v1/timeline-1.0/render | $0.10 per ceil(output minute) |
| Reusable audio join or split | POST /v1/timeline-1.0/audio | $0.01 per job |
The arithmetic for one 30-second episode
Take one trim, one caption job and one timeline render. The timeline reserve is ceil(audio.duration_seconds / 60) minutes, so a 30-second episode reserves one minute. That is $0.02 + $0.20 + $0.10 = $0.32 of finishing, before the Format run itself.
The Format run is billed against the cap you set. Read usage.billable_amount_usd_micros on the terminal receipt to see what it spent. Add that to the $0.32 to get the cost of the episode.
Keep the chain repeatable
Every write in the chain needs an Idempotency-Key. Derive each one from the episode and the step, for example ep-14-trim and ep-14-captions. A retry of a step replays the original job and does not charge again.
The finishing steps read only this workspace's media.sume.com URLs (captions is the exception and takes a public HTTPS video URL). That is why the Format's output URL feeds them directly.
- Check a timeline before you pay with
POST /v1/timeline-1.0/plan. It is unbilled and returnsbillable_minutesandestimated_cost_usd_micros. - Check a dim or crop with
POST /v1/video-filter/check. It is free. - Poll each job at
GET /v1/jobs/:id/status, then readGET /v1/jobs/:id/result.
When a new model launches mid-series
Only the first step changes. Fork the Format, point its recipe at the new model, and run one episode with a low cap. The trim, captions and timeline calls stay as they are, because they work on any MP4 that lives on the Sume media host.
That is the reason to put the finishing steps outside the Format: the part that changes with every model launch stays small, and the part that must stay stable does.
Related posts
More in Use cases
- What is a critical error in TTS? A rubric after Nova 2 Sonic's 28%
Amazon reports 28% fewer critical errors in Nova 2 Sonic on an internal set and does not define them. Define yours, then tally with a read-back on Sume.
- What counts as original for an AI-made YouTube Short?
YouTube's Oct 2026 Shorts update favors original work. Here is what its own pages say about AI, templates and edits, and what they leave unsaid.
- What size should a TikTok video be? 2026 ad minimums by type
TikTok's own ad pages list 9:16 at 540x960 minimum for in-feed and 720x1280 for App Bundle. See the table and the Sume parameters that hit each one.
- Which AI video model for ads, product shots or talking heads?
On Sume, pick Wan 3.0 or Omni for ads, Kling 3 for silent product shots, and the Avatar Video route for talking heads. One table with rates and limits.
Written by Sume