AI thumbnail or real frame for TikTok? The label angle, and a pick

The simplest cover is a real frame from the clip, and Sume can pull one. A search-result snippet from Cross Globe Marketing (read 2026-10-02) reports that AI-assisted thumbnails need no TikTok label, but it is only a snippet, and a fully generated cover is a different thing from an assisted one. If you avoid generating the cover, the question mostly goes away.

How do you pick a real cover frame?

Video frames extracts stills from one Sume-hosted clip at times you name, as durable image files at source size, and it is unbilled. Give it at with 1 to 24 times, or fps up to 2. Each time must fall inside the clip, or the worker returns frame_time_out_of_range and names the duration.

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

When would you generate a cover instead?

When no frame works: the face is mid-blink, or the key product is out of shot. Images can generate a still with aspect_ratio 9:16. Treat that still as AI content, because it is a photorealistic image made by a model, which one summary lists as a label trigger.

Does the cover have to match the video?

Yes in practice: a cover that promises a scene the video lacks reads as misleading, labeled or not. Pull the frame from the clip and you cannot make that mistake.

Pick two or three candidates, view them at phone size, and choose the one where the face or product reads at a glance. Then set the cover in TikTok's own upload flow.

What does the frame job cost and return?

Video frames is unbilled, and a successful submit returns 202 with a request id that is the job id. When the resource is ready, frames lists the durable image URLs with time, width and height. One failed instant comes back with a null URL and does not fail the whole job. The source clip can be up to 300 seconds.