First Draft cuts one Reel at a time: rough-cut 20 clips by API

Instagram's First Draft runs inside the app on the clips you picked. For a batch of 20 talking clips, trim and join them by API: cost table, steps, one curl.

5 min readSume
All posts

Instagram's First Draft is a one-Reel tool: you pick clips inside the app, and it assembles a starting edit for those clips. If you have 20 raw talking clips to turn into a week of Reels, the repeatable route is an API: read a transcript with sentence segments, choose the ranges to keep, trim each range, and join the keepers in one timeline render.

TechCrunch reports that First Draft takes the clips you already selected, cuts pauses and trims them, and returns a first pass in under ten seconds, rolling out on iPhone first and reachable from the camera or from the Reels gallery (TechCrunch, read 2026-10-07). That is a good fit for one phone and one Reel. It is not a queue.

What a batch needs that an in-app tool does not give you

A batch has three needs: the same decision rule on every clip, a record of what was cut, and a way to rerun one clip without redoing the rest. Sume gives you those as separate jobs, each with an idempotency key, so a retry never renders twice.

  • Decision rule: the same silence_split_seconds value on every inspect call, so a pause means the same thing in clip 3 and clip 17.
  • Record: each trim returns a new artf_ URL and actual_start_seconds; the source is never changed.
  • Rerun: change one range, resubmit one trim with a new idempotency key, and re-run the join.

The route, step by step

Every Sume media job reads a clip that already sits on media.sume.com in your workspace, so the first step is POST /v1/media-imports for each source file. Then:

  • POST /v1/video-inspect with transcribe: true and segmentation: { mode: "sentence", silence_split_seconds: 0.6 }. The range allowed is 0.2 to 3 seconds. You get sentence segments to choose from.
  • Pick the segments to keep in your own code (drop false starts, keep the take that ends cleanly).
  • POST /v1/video-trim once per kept range, with start and end or duration. Default precision is exact, a frame-accurate re-encode.
  • POST /v1/timeline-1.0/render to join the trimmed clips for one Reel, with a short fade transition if you want one.
curl -X POST https://api.sume.com/v1/video-trim \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: take-07-keep-1" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/take-07.mp4",
    "start": 1.4,
    "end": 9.8
  }'

What it costs for 20 clips and 4 Reels

The prices below are the public rates in the Sume docs on the read date. Inspect with a transcript adds Modal compute on top of the speech-to-text rate, and the docs bill that compute at cost plus margin, so the total is a floor, not a quote. Confirm live prices in GET /v1/catalog.

Sume public rates for a 20-clip, 4-Reel batch (read 2026-10-07)
StepRateCountSubtotal
Speech-to-text in inspect$0.01 per audio minute20 clips x 1 min$0.20 plus compute
Video trim$0.02 per job20 trims$0.40
Timeline render$0.10 per started output minute4 Reels of 60 s or less$0.40
Total floor$1.00 plus inspect compute

What to prepare before you post

Keep the raw clips. First Draft is described as a starting edit, and the same rule applies to an API cut: a human watches the first frame and the last frame of every Reel before it goes out. Pull both with video frames if you want to check without opening the file.

If you only have one or two clips a week, the in-app tool is faster. The API route earns its keep when the same cut rule has to run on a folder.

Where the API route is weaker

The route is a rough-cut route, and it is blunter than an in-app tool in three ways. It has no visual preview until the job finishes, so you review the output rather than the plan. It does not choose between two takes for you; your code applies a rule. And it needs your clips on media.sume.com before it can read them, which is one more step than opening the camera roll.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume