Board game trailer: table clips and box art as Gemini Omni references

For a board game trailer, trim three shots of your own table to 3 seconds or less each, upload them with the box art as references, and send them to gemini-omni-flash-1.1. The model accepts up to 3 reference_video_urls (each 3 s or shorter) and up to 10 reference_image_urls, and you address them in the prompt as <VIDEO_REF_0> and <IMAGE_REF_0>. The result is 3 to 10 seconds with synced audio.

Source: Video Router, read 2026-10-04.

Prepare the references

The 3-second cap is per reference clip, not per request. Cut each reference to the one gesture you want the model to copy: a die roll, a card flip, a hand moving a meeple.

Trim, then reference

Use Video trim to cut a long phone clip down: start, end and audio of keep or drop, a flat $0.02 per trim. Keep precision: "exact" (the default) when a card flip must land on the frame you chose; keyframe is a stream copy that may start a GOP early.

Request

Write the prompt as a shot list in one paragraph and name which reference supplies what. Do not ask the model to print the game's title. Add the title and release date as a plate or caption afterward.

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: boardgame-trailer-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Trailer for a cooperative board game. Hands roll a die as in <VIDEO_REF_0>, flip a card as in <VIDEO_REF_1>, slide a wooden token across the map from <IMAGE_REF_0>. Warm lamp light, a low drumbeat that builds.",
    "reference_video_urls": ["https://example.com/roll.mp4", "https://example.com/flip.mp4"],
    "reference_image_urls": ["https://example.com/map.jpg", "https://example.com/box.jpg"],
    "resolution": "1080p",
    "aspect_ratio": "16:9",
    "duration": 10,
    "mode": "async"
  }'

Build the trailer in Timeline

One clip is not a trailer. Generate two or three, then join them in Timeline 1.0 with wipeleft or slideup transitions of 0.5 seconds or less, and a music bed as soundtrack. A fade chain longer than 8 is refused, which is not a concern for three clips.