Descript scrubbing 2-5x faster vs a video-frames contact sheet

Descript's changelog says scrubbing is 2 to 5 times faster because it pre-computes video chunks. That helps when you move through a clip by hand. If you only need to see what is in a file without opening an editor, Sume's POST /v1/video-frames returns 1 to 24 stills at times you choose, or sampled at up to 2 frames per second, as durable image URLs.

How the frames route works

Pass exactly one of at[] (seconds) or fps. format is jpeg by default or png, and max_edge clamps the long side between 16 and 2160. The source can be up to 300 seconds. The call always returns 202; read the resource later at GET /v1/video-frames/{id}, and a failed instant comes back with a null url without failing the job. Billing is by Modal compute, so check the catalog rather than assuming a flat rate.

curl -X POST https://api.sume.com/v1/video-frames \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: contact-sheet-001" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_demo/talk.mp4","at":[0,5,10,15],"max_edge":512}'

Using the stills

The stills can go to a reviewer, or to a vision model for scene notes. An instant at or past the clip duration fails as frame_time_out_of_range, so probe duration first with a free video-inspect call.

Honest limits

• It is not a scrubber: you get stills, not playback.
• Sources longer than 300 seconds are not accepted.
• Descript's speed gain is a vendor claim; it was not measured here.