Quinceañera invitation teaser: portrait and venue as references

For a quinceañera invitation teaser, send the honoree's portrait, the dress, the venue and the invitation art as reference_image_urls (up to 10) to gemini-omni-flash-1.1, and point at each one in the prompt as <IMAGE_REF_0>, <IMAGE_REF_1> and so on. You get a 3 to 10 second vertical clip with synced audio, and the date and address go on afterward as a plate or captions, not inside the generated video.

Facts are from the Video Router and Video generation pages, read 2026-10-04.

How the references work

Reference-to-video is one of four request shapes Gemini Omni Flash 1.1 routes by. The reference list is addressed by position, starting at 0.

Submit the teaser

List the images in the order you will mention them. Four are enough for most invitations: a clear portrait first, then the dress, then the venue exterior, then the invitation card for color and type.

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: quince-teaser-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "A vertical invitation teaser. The young woman in <IMAGE_REF_0> turns toward camera in the dress from <IMAGE_REF_1>, in the garden courtyard from <IMAGE_REF_2>. Warm evening light, petals drifting, the color palette of <IMAGE_REF_3>.",
    "reference_image_urls": [
      "https://example.com/portrait.jpg",
      "https://example.com/dress.jpg",
      "https://example.com/venue.jpg",
      "https://example.com/invitation-card.jpg"
    ],
    "resolution": "1080p",
    "aspect_ratio": "9:16",
    "duration": 10,
    "mode": "async"
  }'

Put the details on afterward

Do not ask the video model to render the date, time or address. Make the details as a still plate in your design tool, import it, and put it on the clip with Timeline compose in overlay mode (position, width_ratio 0.05 to 1, margin_ratio). Compose is a flat $0.02 job and the clip's length sets the output length.

If the invitation is bilingual, the cheapest safe route is the same: one plate per language, one compose job each, same teaser underneath.

Likeness note

Reference images are likeness inputs. Use photos of the honoree that her family has agreed to share and tell guests when a clip is AI-generated. Sume's docs do not describe a consent check, so that step is yours.