One video ad into many variants with the same pacing, by API

Run one edit job per variant. Sume's video router edit takes a single source clip as video_url plus a prompt that describes the change, for example "Replace the bottle with an apple. Keep everything else the same." Repeat that with one change per job and you get a set of variants of one ad. Nothing here is a single "multiply" call; you write the loop.

Higgsfield's changelog (Aug 25, 2026) describes its Ad Multiplier as turning one video ad into many variants with the same video, edit and pacing, and new characters, clothing, locations and objects. Sume facts are from Video router and Video generation, read 2026-10-01.

What does one variant request look like?

A video_to_video edit takes video_url; the prompt describes the edit, and resolution is optional (default 720p). There is no aspect_ratio or duration. video_url is the edit source, not a reference, so it cannot be combined with image_url, end_image_url or reference_*_urls.

const edits = [
  "Replace the bottle with an apple. Keep everything else the same.",
  "Move the scene to a beach. Keep everything else the same.",
];
for (const [i, prompt] of edits.entries()) {
  const res = await fetch("https://api.sume.com/v1/video-router/generate", {
    method: "POST",
    headers: {
      Authorization: "Bearer " + process.env.SUME_API_KEY,
      "Content-Type": "application/json",
      "Idempotency-Key": "ad-variant-" + i,
    },
    body: JSON.stringify({
      model: "gemini-omni-flash-1.1",
      prompt,
      video_url: "https://example.com/clip.mp4",
      resolution: "720p",
      mode: "async",
    }),
  });
  console.log(i, (await res.json()).request_id);
}

What should I keep fixed between variants?

What does each variant cost?

Each is its own paid job. The docs say billing is reserved on submit at provider list × 1.25, and that video models accept video_url as an input reference; check the per-model fields in the catalog before choosing a model. Multiply the one-clip cost by your variant count before you submit, and see hook variations for UGC ads for a related way to vary the opening.

Does the edit really preserve the pacing?

The Sume docs only promise that the prompt describes the edit; they do not guarantee identical cuts. Review each output against the source, and rerun a variant with a narrower prompt if something else changed.