Prove an AI clip still shows the right color and size

TikTok Shop's AI policy prohibits altering a product's size, color, material, quantity, features or safety attributes, and prohibits changing its appearance so it does not match what customers receive. The check is simple: take stills from the finished clip and set them beside the listing photo. Sume's video frames API gives you up to 24 stills per call.

What is prohibited, and what is permitted

Extract the stills

Video frames reads one media.sume.com clip (import it first) and needs exactly one of at[] (1 to 24 times, each from 0 up to the clip duration) or fps (up to 2, capped at 24 frames). The source can be up to 300 seconds. Frames are jpeg by default or png, and max_edge (16 to 2160) clamps the long edge; leave it off to keep the source size. A submit returns 202; read the result when resource_status is ready.

curl -X POST https://api.sume.com/v1/video-frames \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: color-check-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
    "at": [0.5, 2, 4, 6, 8]
  }'

Compare against the listing

Open each still next to the listing photo and check one attribute at a time. The tool extracts frames; it does not judge them, so the comparison is a human or your own tooling.

• Color: same shade under similar light, not a restyled one.
• Size: nothing in frame implies a different scale.
• Material: finish and texture match (matte stays matte).
• Quantity: the count on screen is what the buyer receives.
• Features: no button, port or accessory that the product lacks.

Make it routine

Pick the stills the same way every time: the opening frame, the first product close-up, the middle of the demo and the final frame. A fixed rule keeps two reviewers from picking different frames and disagreeing about the result.

Keep the stills with the clip. If a buyer disputes how a product looked, a saved comparison of listing photo and clip stills shows what you checked and when.

Cost and scope

Video frames is billed by its Modal compute (container seconds x Modal list x 1.25, plus the platform fee), and the bill is never more than the hold. The docs do not publish a flat price for it, so check GET /v1/catalog instead of assuming one. A failed instant leaves that frame's url as null without failing the job, so count the returned URLs.