2-second YouTube intro sting: Wan 3.0 has the lowest floor

For a 2-second YouTube intro sting, use wan-3.0. The Video Router docs give it a 2 to 30 second window, the lowest floor among the models listed, while seedance-2.5 accepts 4 to 30 and gemini-omni-flash-1.1 accepts 3 to 10. For a sting shorter than 2 seconds, generate a longer clip and cut it with video trim.

Source: Video Router and Video generation, read 2026-10-04. Always read capabilities from GET /v1/video-router/models, which the docs treat as the source of truth.

Duration windows

Limits differ per model, so a failed request often means a length outside the model's window rather than a wrong prompt.

Request

A sting is mostly a logo reveal, so give the model a clean still. Use /v1/videos, which accepts frame_images for a first frame, and ask for one move.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan-3.0",
    "prompt": "A glass shard logo assembles from light streaks on a dark background and settles. One camera push, no text changes.",
    "frame_images": [
      { "type": "image_url", "image_url": { "url": "https://example.com/logo-final.png" }, "frame_type": "last_frame" }
    ],
    "duration": 2,
    "resolution": "720p",
    "aspect_ratio": "16:9"
  }\

Check the frame support first

Whether wan-3.0 accepts a last_frame entry is a capability you should check in the model list (supported_frame_images) before relying on it. If it does not, use first_frame or a prompt only.

Shorter than the floor

When you need 1.2 seconds, trim. Video trim cuts [start, end) into a new MP4 for a flat $0.02, with a 0.2 second minimum cut, and precision: "exact" re-encodes frame-accurately. The audio can be kept or dropped.