Amazon video: key benefit in the first 10 seconds, checked in 5 stills

Pull five stills at 1, 3, 5, 7 and 9 seconds with POST /v1/video-frames and check that the product and its main benefit are on screen by the third or fourth one. Amazon's seller page (read 2026-10-08) says the most successful shoppable videos highlight the most important feature in the first 5 to 10 seconds and use basic text overlays so the message works when watched without sound.

What Amazon says

The page recommends 30 to 90 seconds and says the strongest examples state the key benefit within the first 5 to 10 seconds. It says to include simple text overlays for important information so customers still get the message without sound.

It does not say Amazon measures this. It is advice, not a review criterion, so a clip that misses it can still be approved.

The five-still check

POST /v1/video-frames takes one clip that is already stored on media.sume.com and an at array of 1 to 24 times. Each time must be at least 0 and below the clip duration, otherwise the worker fails with frame_time_out_of_range. The response returns durable image URLs, and a frame that fails to extract has a null url without failing the job.

Sume bills the job by its Modal compute at list times 1.25 plus the platform fee, and the docs give no fixed amount, so read the reserve from the submit response.

Read the stills in order, as a viewer would, and ask one question of each: could someone who has never seen the product say what it does? If the third still is still a title card or a pack shot, the first 5 to 10 seconds are being spent on setup, which is the opposite of what Amazon recommends.

curl -X POST https://api.sume.com/v1/video-frames \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: amazon-first-10s-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/overview.mp4",
    "at": [1, 3, 5, 7, 9],
    "max_edge": 480
  }'

If the check fails

Re-render the opening only. Generate a new first clip that starts on the product and states the benefit, then join it to the existing middle and end on a Sume timeline instead of regenerating the whole video.

A 10-second opening on Wan 3.0 at 720p costs $1.25 per take, so three takes cost $3.75. That is cheaper than redoing a 90-second video at $11.25.

Run the same five-still check on every video in a batch before you submit. Amazon reviews each upload, and at holiday volume a resubmission can cost up to seven business days, so a cheap check beats a rejected review.

• Import the clip with POST /v1/media-imports first; the frames API refuses other hosts.
• Use max_edge to keep the stills small; it accepts 16 to 2160.
• Overlay text can be burned with the caption endpoint and cues.