First Draft only cuts clips you have: generate the b-roll first

First Draft assembles clips you have already selected, so any shot that is not in your camera roll has to exist before you press it. If a Reel needs a cutaway you did not film, generate it as a 9:16 clip first: Sume's Video Router takes seedance-2.5 for 4 to 30 seconds or gemini-omni-flash-1.1 for 3 to 10 seconds, both in 9:16.

TechCrunch says First Draft works from selected clips, in the camera or the Reels gallery (TechCrunch, read 2026-10-07). The tool does not invent footage, so planning the shot list is your job.

Which model for which cutaway

ByteDance describes Seedance 2.5 as generating up to 30 seconds per pass with audio and video produced together, released 2026-07-31, with API access listed as coming via BytePlus ModelArk at the time of its post (ByteDance Seed, read 2026-10-07). Sume's docs list the model as available now with the limits below.

One call

The request below is the shape from the Video Router page. The job is asynchronous: you get a job id, poll GET /v1/jobs/:id/status, then read GET /v1/jobs/:id/result for the artifact URL on media.sume.com.

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: broll-coffee-01" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "Vertical close-up of espresso pouring, natural light",
    "resolution": "720p",
    "duration": 6,
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

What to prepare

Price comes from the model list, not from this page: GET /v1/videos/models returns pricing_skus per model, and Sume bills at its house margin on top of the provider rate. Check it before you queue ten cutaways.

Keep cutaways short. A 4 to 6 second shot between two talking clips is enough for a Reel, and shorter jobs are cheaper to regenerate when the first result is wrong. Then save the generated clips to your phone or send them to your editor, and let First Draft or your own trim do the cut.

Writing a cutaway prompt that cuts well

A cutaway has one job: cover a jump or add context for two to six seconds. Write it like a shot list line, not a story. Say what is in frame, the camera move, and the light, and keep it to one idea. A prompt such as a vertical close-up of espresso pouring in natural light gives the model less to get wrong than a scene with three actions.

Ask for 9:16 in the request, not in the prompt, with aspect_ratio: "9:16". For a clip you plan to cut under a voice, you do not need the model's sound, so check whether the model you pick always generates audio. The docs note that gemini-omni-flash-1.1 has native synced audio always on, and you can mute or replace it in the timeline.