TikTok Ad Teardown Before You Remake It: Trending Search and Ingest

The short answer

Search Sume's trending videos endpoint for ranked public TikTok metadata ($0.10 per call), then read the clip you will study with reference ingest, which returns frame-exact shots, on-screen text, audio facts and a labeled strip for one clip of up to 300 seconds. Reference ingest is dest first: Sume lists it only where its feature flag is on (development auto-on, production opt-in), so check tools_list or the docs before you plan on it in production.

Step 1: search

POST /v1/trending-videos/search takes a query (required in production), window (yesterday, this-week, this-month, last-3-months, last-6-months, all-time), limit of 1 to 50 and an optional two-letter region. Results have the canonical watch URL, a cover thumbnail, author handle, metrics and relevance. They do not include a downloadable video, and the docs say the endpoint does not return source files for face swap or captions.

So the research output is a list of links and numbers. To study a clip's structure you need a copy you have the right to use, such as your own ad or a clip a creator gave you, imported into Sume.

Step 2: read the clip

Reference ingest returns shots[] with frame-exact cuts, text_tracks[] read at source resolution, audio facts (a silent gate, speech presence and beats for music) and uncertain[] for lines that need another look. Its purpose field accepts reference_remix, brief_format, face_swap or qa, which Sume stores but does not interpret. For reference_remix, the read transcribes by default and adds the speech-to-text rate; set speech.allow_billed_stt to false to opt out.

Use the manifest as a brief: the number of cuts in the first 3 seconds, where the text appears, whether the sound is speech or music. Then write your own script. Copying another advertiser's footage, voice or distinct creative is not what this is for.

Budget the research at a few dollars, not a few cents: ten searches are $1.00, and a cached repeat browse can be served without another upstream call. Keep a short log of the query, window and region for each run, since a result that changes next week was probably the window moving, not the idea failing. Then check the platform's own ad specs before you render, as in the TikTok upload checklist.

• Clips over 300 seconds are refused with source_too_long_for_reference_ingest; use video inspect instead.
• A clip with no video stream gives source_no_video_stream.
• Production still requires the query on trending search; the browse feed with no query is on for the dev API.