Swap the person, then the product: Recast and Omni edit in turn

A winning product ad usually needs two changes for a new market: a different presenter and a different product. No single Sume model does both, but two jobs in a row do. H3 Max Recast replaces the people from reference photos and keeps motion, camera and cuts (fal's page, read 2026-10-04). Gemini Omni Flash 1.1 then edits the result with a prompt.

The split matters because the two models take different inputs. Recast wants one photo per person. Omni edit takes only video_url and a prompt, and Sume's Video Router docs say you cannot send video_url together with image or reference fields. So the person comes from a photo, and the product has to be described in words.

Step one: the person

Submit h3-max-recast with the source video, the presenter photo and duration equal to the inspected source length. An 8 second clip at the default 768p bills $3.00 on Sume ($0.375 a second). Check the result before you move on; there is no point editing a swap you would reject.

Step two: the product

Send the finished clip to gemini-omni-flash-1.1 as video_url, with the instruction in prompt. Sume's own example is "Replace the bottle with an apple. Keep everything else the same." The docs list Omni at 3 to 10 seconds, so trim a longer source first. Omni's native audio is always on, and it bills the provider list price times 1.25 for each output second.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: swap-product-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Replace the red mug with a white ceramic teapot. Keep the presenter, motion and camera the same.",
    "video_url": "https://example.com/recast-result.mp4",
    "resolution": "720p"
  }'

Check both swaps with stills

A prompt edit can leave a trace of the old product in a frame or two. Sample the final clip with video frames at the seconds where the product is on screen, up to 24 stills per call, and look for the old item and for the new one at the same size. The earlier post on checking a product swap with video frames walks through the sampling.

Because the clips must be 5 to 10 seconds to pass both stages, this chain suits short hooks and cutdowns, not a 30 second spot. For a longer ad, recast the full clip and re-shoot the product beat separately.