Game concept art to a teaser clip: Omni image refs, 8 s for $1.00

The short answer

To turn game concept art into a teaser, pass up to 10 images as reference images to gemini-omni-flash-1.1 and cite them in the prompt as <IMAGE_REF_0>, <IMAGE_REF_1> and so on. An 8-second 720p clip bills $1.00 on Sume (8 seconds at $0.125), and 1080p bills $1.50. The model always generates native audio, and clips run 3 to 10 seconds.

How the reference tags work

A concept board usually has a hero character, an environment and a prop. Omni treats each image as a reference rather than a frame to start from, so the video is not locked to any one picture. The tags are 0-based and follow the order of your list: the first image is <IMAGE_REF_0>, the second is <IMAGE_REF_1>. The reference media is sent before the prompt, so write the prompt as if the tags already stand for the pictures.

Request

The prompt names each reference by role.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: teaser-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "<IMAGE_REF_0> crosses the ruined bridge in <IMAGE_REF_1>, wind in her cloak, low sun",
    "duration": 8,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/hero.png"}},
      {"type": "image_url", "image_url": {"url": "https://example.com/bridge.png"}}
    ]
  }'

Format and limits

Omni accepts 16:9 and 9:16 only, so a vertical trailer for a store listing is a 9:16 request and a cinematic one is 16:9. Reference videos are also allowed (up to 3, each up to 3 seconds), tagged <VIDEO_REF_0>, which is useful for borrowing a camera move.

Price of an 8 second teaser

Billing is per output second by resolution. The reference images do not add a separate charge in the Sume catalog row.

Draft at 360p, then pay for the keeper

Run the 360p draft first. At $0.30 an 8-second clip is cheap enough to try five prompts, then spend $1.00 or $1.50 on the one that works. If you want MiniMax H3 instead, it takes up to 9 reference images, but its catalog row notes that images past the first five carry a $0.08 list surcharge each, so keep the board to five.