One ad into six variations: hooks, crops, captions, $0.12 of trims

Meta's Laura Geller case study describes using AI to turn one ad into six variations with new hooks, crops and captions. With Sume you can do the mechanical half of that yourself: three different start points times two output sizes is six POST /v1/video-trim jobs, and trim is $0.02 per job, so $0.12 for the set before captions.

Meta facts are from its Business news post, read 2026-10-01. Sume facts are from Video trim and Video captions. This post does not reproduce Meta's tooling or its results.

How do the six break down?

A hook is the first second or two, so a new hook means a new start. A crop is a new output width and height. A caption is a separate pass over the finished clip.

What does one cut look like?

Import the master first, since trim only reads media.sume.com clips, and send a distinct Idempotency-Key for each variant. Check the framing of every crop; do not assume a 16:9 master turns into a good 9:16 frame.

curl -X POST https://api.sume.com/v1/video-trim \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: variant-b1" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/master.mp4",
    "start": 4,
    "duration": 15,
    "precision": "exact",
    "output": { "width": 1080, "height": 1920 }
  }'

How do I add captions?

Run POST /v1/video-captions on each finished clip, with a video_url and an optional style such as slam or punch. Captions need audible speech, or you pass authored cues instead. Write a different line per hook so the variants are really different. Which one wins is a test you run in Meta's tools, not something Sume predicts.