Remix a viral TikTok video with AI: what Sume does and does not do

To remix a viral TikTok video with AI on Sume, you research with trending search, then work from a clip you already have the right to use. Trending search returns public TikTok metadata, not video files: the Trending videos page, read 2026-09-29, says it does not download or mirror videos and does not return raw source files for later steps.

That shapes the workflow. Sume does not fetch a TikTok for you to remix; you bring your own clip into your workspace, and the tools read it and help you make something new.

What does trending search give me?

POST /v1/trending-videos/search returns ranked entries with the public watch url, an optional cover_url thumbnail, the caption, author handle, metrics and relevance. In production a query is required and limit runs 1 to 50. One call is $0.10. Use it to see what formats, hooks and lengths are working for a brand or keyword.

How do I read a reference clip?

Import a clip you own or are licensed to use, such as your own past ad or a licensed stock clip, and run POST /v1/reference-ingest on it. You get its shots, on-screen text, audio facts and a labeled strip. It is dev-first and opt-in in production, so check that it is listed for you. See shot breakdown.

How do I make new footage from it?

• Do not copy another creator's footage, voice or on-screen claims. Recreate the structure (hook, cuts, pacing) with your own product and assets.
• Audio and video references are honored by the Seedance 2.x models, Wan 3.0, MiniMax H3 and H3 Max; check supported_input_references from GET /v1/videos/models for the model you pick.

Is this allowed on TikTok?

Sume's docs do not state platform policy. TikTok's own rules on AI-generated content and on copying others' videos apply to what you post; does TikTok allow AI-generated content covers the labeling side.

What should the new brief keep?

Keep the structure, not the material.

• Shot count and timing: the length of each entry in shots[].
• Text beats: which lines appear when, from text_tracks[], rewritten for your product.
• Audio plan: if audio.silent is true, plan new music instead of reusing the source audio.
• Your own product photos and voice, passed as frame_images or input_references to a generation call.