Turn a TikTok trend into an AI video for my product: 4 steps

Search the trend, read a clip you own, rebuild its beats around your product photo, then generate 9:16. The four Sume calls in order and what each returns.

4 min readSume
All posts

To turn a TikTok trend into an AI video for your product, run four steps: search the trend for research, read a clip you own with reference ingest, write your own beat list from its shots, then generate the new video with your product image. Only step one touches TikTok data, and it returns metadata and watch links, not video files.

That is why steps two to four use a file you already have. Everything below comes from the Sume docs pages cited at the end, read 2026-09-29.

What are the four steps and calls?

Trend-to-video steps and the Sume call for each, read 2026-09-29.
StepCallYou get
1. Find the trendPOST /v1/trending-videos/searchRanked TikTok metadata, $0.10 per call
2. Read a clip you ownPOST /v1/reference-ingestShots, text lines, audio facts (unbilled)
3. Write your beatsYouOne line per shot, with your product
4. GeneratePOST /v1/videosA job that returns the clip

How do I narrow the trend search?

Pick a window from yesterday, this-week, this-month, last-3-months, last-6-months or all-time, and optionally a two-letter region. Production requires a query (a brand, product, creator or keyword up to 200 characters) and takes limit 1 to 50. platform supports only tiktok for now.

How do I turn shots into a brief?

Ingest a clip in your workspace and copy the structure, not the material: how many shots, how long each runs, where text appears. If audio.silent is true the docs say to plan new music and discard the source audio. Reference ingest is on in development and opt-in in production, so confirm it is listed for you first.

How do I put my product into the new clip?

Pass your product photo as frame_images with frame_type first_frame for image-to-video, or as input_references for reference-to-video, where the model uses it as visual guidance rather than an exact frame. If both are sent, frame_images wins. Set aspect_ratio to 9:16 for a vertical clip.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: trend-remix-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "Shot 1: hand lifts the bottle into frame. Shot 2: close-up of the label.",
    "aspect_ratio": "9:16",
    "duration": 8,
    "frame_images": [{
      "type": "image_url",
      "image_url": { "url": "https://media.sume.com/artifacts/example/bottle.png" },
      "frame_type": "first_frame"
    }]
  }'

What should I leave out?

Do not reuse another creator's footage, voice or claims; write your own. Sume's docs do not state TikTok's policy, so check TikTok's own rules for what you post, and see the fuller remix workflow.

What does it cost to run the loop once?

One trend search is $0.10. The reference ingest manifest is unbilled, so reading a clip you own adds nothing unless you opt into the transcript. The generation call is priced by the model you pin; the Video generation docs say billing is reserved on submit at provider list times 1.25 on every model, so read the price for your model and duration in the catalog before you submit.

Repeat browse calls to trending search may be served from an in-process cache, so re-running the same research does not always re-hit the upstream data source. That is a note about cost, not a promise about freshness.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume